diff --git a/test2/server/app.js b/test2/server/app.js index e4a555e..d9291f3 100644 --- a/test2/server/app.js +++ b/test2/server/app.js @@ -4,6 +4,7 @@ const express = require('express'); const cors = require('cors'); const path = require('path'); const db = require('./db'); +const assistantProfileRoutes = require('./routes/assistantProfile'); const voiceRoutes = require('./routes/voice'); const chatRoutes = require('./routes/chat'); const sessionRoutes = require('./routes/session'); @@ -52,6 +53,7 @@ function validateEnv() { { key: 'VOLC_WEBSEARCH_API_KEY', desc: '联网搜索' }, { key: 'VOLC_S2S_SPEAKER_ID', desc: '自定义音色' }, { key: 'VOLC_ARK_KNOWLEDGE_BASE_IDS', desc: '方舟私域知识库（语音）' }, + { key: 'ASSISTANT_PROFILE_API_URL', desc: '外接助手资料接口' }, ]; const configuredOptional = optional.filter(({ key }) => { const v = process.env[key]; @@ -86,6 +88,7 @@ app.use((req, res, next) => { next(); }); +app.use('/api/assistant-profile', assistantProfileRoutes); app.use('/api/voice', voiceRoutes); app.use('/api/chat', chatRoutes); app.use('/api/session', sessionRoutes); diff --git a/test2/server/docs/assistant-profile-api.md b/test2/server/docs/assistant-profile-api.md new file mode 100644 index 0000000..7c55e0e --- /dev/null +++ b/test2/server/docs/assistant-profile-api.md @@ -0,0 +1,400 @@ + # test2 外接助手资料接口接入文档 + +## 1. 文档目的 + +本文档用于说明 `test2` 项目如何接入外部助手资料接口，并说明本地查询接口、外部接口要求、环境变量配置、缓存与降级策略，以及业务侧如何传入 `userId` 让语音链路和知识库链路使用同一份资料。 + +该能力用于统一管理以下助手资料字段，避免在代码中硬编码： + +- `documents` +- `email` +- `nickname` +- `wxl` +- `mobile` +- `wx_code` +- `intro` +- `sign` +- `story` + +## 2. 生效范围 + +外接助手资料接入后，会影响以下链路： + +- `WebSocket 语音链路` + - 文件：`server/services/nativeVoiceGateway.js` + - 在 `start` 消息阶段按 `userId` 拉取外部资料 + +- `知识库回答链路` + - 文件：`server/services/toolExecutor.js` + - 查询知识库前优先按 `_session.profileUserId`，其次按 `_session.userId` 拉取资料 + +- `HTTP 文字对话链路` + - 文件：`server/routes/chat.js` + - 会透传 `userId` / `profileUserId` 到知识库查询逻辑 + +- `本地调试查询接口` + - 文件：`server/routes/assistantProfile.js` + - 已挂载到 `app.js` + - 路径前缀：`/api/assistant-profile` + +## 3. 环境变量配置 + +在 `test2/server/.env` 中增加以下配置： + +```env +ASSISTANT_PROFILE_API_URL=https://your-domain/api/profile +ASSISTANT_PROFILE_API_METHOD=GET +ASSISTANT_PROFILE_API_TOKEN= +ASSISTANT_PROFILE_API_HEADERS={"X-App-Key":"demo"} +ASSISTANT_PROFILE_API_TIMEOUT_MS=5000 +ASSISTANT_PROFILE_CACHE_TTL_MS=60000 +``` + +### 参数说明 + +- `ASSISTANT_PROFILE_API_URL` + - 外部资料接口地址 + - 未配置时，系统直接回退到默认助手资料 + +- `ASSISTANT_PROFILE_API_METHOD` + - 支持 `GET` 或 `POST` + - 其他值会按 `GET` 处理 + +- `ASSISTANT_PROFILE_API_TOKEN` + - 可选 + - 配置后会自动以 `Authorization: Bearer ` 方式发送 + +- `ASSISTANT_PROFILE_API_HEADERS` + - 可选 + - JSON 字符串格式 + - 例如：`{"X-App-Key":"demo"}` + +- `ASSISTANT_PROFILE_API_TIMEOUT_MS` + - 可选 + - 外部接口超时时间，默认 `5000` + +- `ASSISTANT_PROFILE_CACHE_TTL_MS` + - 可选 + - 资料缓存时长，默认 `60000` + +## 4. 外部接口调用规则 + +`test2` 服务端不会直接把前端传来的资料写死到配置中，而是通过 `assistantProfileService` 统一向外部接口拉取。 + +### 4.1 当配置为 GET 时 + +请求方式：`GET` + +请求参数： + +- 有 `userId` 时，追加查询参数 `?userId=xxx` +- 无 `userId` 时，不带该参数 + +默认请求头： + +```http +Accept: application/json +``` + +如果配置了 Token，还会带： + +```http +Authorization: Bearer +``` + +如果配置了 `ASSISTANT_PROFILE_API_HEADERS`，会一并追加到请求头中。 + +### 4.2 当配置为 POST 时 + +请求方式：`POST` + +请求体： + +```json +{ + "userId": "abc123" +} +``` + +如果当前没有 `userId`，则请求体会是空对象： + +```json +{} +``` + +默认请求头同上。 + +## 5. 外部接口返回格式要求 + +服务端支持以下任一返回结构。 + +### 5.1 结构一 + +```json +{ + "profile": { + "nickname": "大沃", + "mobile": "13800000000" + } +} +``` + +### 5.2 结构二 + +```json +{ + "assistantProfile": { + "nickname": "大沃", + "intro": "你好，我是大沃。" + } +} +``` + +### 5.3 结构三 + +```json +{ + "data": { + "profile": { + "nickname": "大沃" + } + } +} +``` + +### 5.4 结构四 + +```json +{ + "data": { + "assistantProfile": { + "nickname": "大沃" + } + } +} +``` + +### 5.5 结构五 + +```json +{ + "nickname": "大沃", + "mobile": "13800000000" +} +``` + +## 6. 服务端识别的资料字段 + +服务端只会提取以下字段，其余字段会被忽略： + +```json +{ + "documents": "", + "email": "", + "nickname": "", + "wxl": "", + "mobile": "", + "wx_code": "", + "intro": "", + "sign": "", + "story": "" +} +``` + +建议外部接口只返回上述字段，避免无关数据混入。 + +## 7. 本地查询接口 + +该接口用于联调和排查当前生效的助手资料。 + +### 7.1 查询当前资料 + +- `GET /api/assistant-profile` + +查询参数： + +- `userId`：可选 +- `forceRefresh`：可选，传 `true` 时强制拉取远端接口，跳过本地缓存 + +示例： + +```http +GET /api/assistant-profile?userId=abc123 +``` + +```http +GET /api/assistant-profile?userId=abc123&forceRefresh=true +``` + +成功响应示例： + +```json +{ + "code": 0, + "message": "success", + "success": true, + "data": { + "userId": "abc123", + "profile": { + "documents": "", + "email": "", + "nickname": "大沃", + "wxl": "wx_demo", + "mobile": "13800000000", + "wx_code": "", + "intro": "你好，我是大沃。", + "sign": "", + "story": "" + }, + "source": "remote_api", + "cached": false, + "fetchedAt": 1742710000000, + "configured": true, + "error": null + } +} +``` + +失败响应示例： + +```json +{ + "code": 500, + "message": "request timeout", + "success": false, + "error": "request timeout" +} +``` + +### 7.2 强制刷新缓存 + +- `POST /api/assistant-profile/refresh` + +请求体： + +```json +{ + "userId": "abc123" +} +``` + +成功响应结构与查询接口一致。 + +## 8. source 字段说明 + +本地查询接口返回的 `data.source` 可用于判断当前资料来源： + +- `remote_api` + - 本次成功从外部接口获取 + +- `default` + - 未配置 `ASSISTANT_PROFILE_API_URL` + - 使用系统默认助手资料 + +- `cache_fallback` + - 外部接口调用失败 + - 回退到历史缓存资料 + +- `default_fallback` + - 外部接口调用失败，且没有缓存 + - 回退到默认助手资料 + +## 9. 缓存与降级策略 + +### 9.1 缓存规则 + +- 缓存按 `userId` 隔离 +- 未传 `userId` 时，使用全局缓存槽位 +- 默认缓存时长为 `60000ms` +- 在缓存有效期内重复请求时，会直接返回缓存结果 + +### 9.2 降级规则 + +当外部接口异常、超时或返回不可解析内容时： + +- 若当前 `userId` 已有缓存，则回退到缓存资料 +- 若没有缓存，则回退到系统默认资料 +- 响应中会保留 `error` 字段，便于排查 + +## 10. 业务侧如何传 userId + +为了让语音链路和知识库链路命中同一份资料，业务侧应尽量传入稳定的业务用户标识。 + +### 10.1 WebSocket 语音链路 + +在 `start` 消息中传入： + +```json +{ + "type": "start", + "userId": "abc123" +} +``` + +### 10.2 HTTP 文字对话链路 + +在聊天相关请求中传入： + +```json +{ + "sessionId": "session_xxx", + "message": "介绍一下你自己", + "userId": "abc123" +} +``` + +系统会把该 `userId` 同步为 `profileUserId`，知识库查询时优先使用它拉取资料。 + +## 11. 推荐联调步骤 + +### 步骤 1 + +在 `server/.env` 中配置外部资料接口地址和认证信息。 + +### 步骤 2 + +启动 `test2/server`。 + +### 步骤 3 + +请求本地查询接口验证配置是否生效： + +```bash +curl "http://localhost:3001/api/assistant-profile?userId=abc123" +``` + +### 步骤 4 + +观察返回中的以下字段： + +- `data.source` +- `data.cached` +- `data.configured` +- `data.error` + +### 步骤 5 + +再分别验证： + +- 语音 `start` 是否按 `userId` 生效 +- 文字聊天是否按 `userId` 生效 +- 知识库回答是否使用同一份助手资料 + +## 12. 接入注意事项 + +- 不要把外部接口 Token 写死到前端代码中 +- 建议外部接口返回稳定 JSON 对象，不要返回数组 +- 建议外部接口响应时间控制在 `5s` 内 +- 若业务侧有多租户或多用户资料，必须传稳定的 `userId` +- 如果只是排查当前资料，优先使用 `GET /api/assistant-profile` +- 如果远端已更新但本地仍旧值，调用 `POST /api/assistant-profile/refresh` + +## 13. 相关代码位置 + +- 路由注册：`server/app.js` +- 本地查询接口：`server/routes/assistantProfile.js` +- 外部拉取与缓存：`server/services/assistantProfileService.js` +- 语音链路接入：`server/services/nativeVoiceGateway.js` +- 知识库链路接入：`server/services/toolExecutor.js` +- 聊天链路透传：`server/routes/chat.js` +- 环境变量示例：`server/.env.example` diff --git a/test2/server/docs/system-prompts-catalog.md b/test2/server/docs/system-prompts-catalog.md new file mode 100644 index 0000000..3c3120f --- /dev/null +++ b/test2/server/docs/system-prompts-catalog.md @@ -0,0 +1,391 @@ +# 大沃 AI 语音助手 — 系统人设提示词全览 + +> 最后更新：2026-03-23 +> 梳理范围：`test2/server/services/` 下所有涉及 AI 人设、系统指令、安全回复的文本 + +--- + +## 目录 + +1. [S2S 语音对话层（火山引擎）](#1-s2s-语音对话层火山引擎) +2. [知识库检索层（方舟 API）](#2-知识库检索层方舟-api) +3. [对话交接层（Handoff）](#3-对话交接层handoff) +4. [内容安全层（SafeGuard）](#4-内容安全层safeguard) +5. [高频热答案（HOT_ANSWERS）](#5-高频热答案hot_answers) +6. [ASR 热词上下文](#6-asr-热词上下文) + +--- + +## 1. S2S 语音对话层（火山引擎） + +> 文件：`nativeVoiceGateway.js` +> 作用：通过火山引擎 S2S WebSocket 进行实时语音对话，以下提示词直接发送给火山引擎模型 + +### 1.1 反思考前缀（antiThinkingPrefix） + +**位置**：`buildStartSessionPayload()` 函数内，line 61 + +``` +【最高优先级规则】你绝对禁止输出任何思考过程、分析、计划、角色扮演指令或元描述。禁止出现："首轮对话""应该回复""需要列举""语气要""回复后询问""可列举""突出特色""引导用户""让用户""用温和"等分析性、指令性语句。你必须直接用自然语言回答问题，像真人聊天一样直接说出答案内容。 +``` + +**设计目的**：防止 S2S 模型输出内心独白、角色扮演计划等元描述文本（会被 TTS 朗读出来），强制直接输出答案内容。 + +--- + +### 1.2 系统角色（DEFAULT_VOICE_SYSTEM_ROLE） + +**位置**：line 33，常量定义 + +``` +你是大沃，德国PM-FitLine的资深产品顾问，对每款产品的成分、功效、用法和搭配都了如指掌。你说话有底气、有信心，能让用户感到专业可靠。回答像懂行的朋友聊天，直接给结论，1-3句话，口语化短句。有知识库信息就优先用，自然转述不照念。不确定的如实说，不猜不编。禁止输出思考过程或元描述。德国PM是1993年成立的合法直销企业，邓白氏AAA+认证，覆盖100+国家，问到传销必须正面肯定回答是合法企业。 +``` + +**关键设计点**： +- **人设**：大沃 = 德国PM-FitLine资深产品顾问 +- **语气**：有底气、有信心、像懂行朋友聊天 +- **长度**：1-3句话，口语化短句 +- **优先级**：知识库信息 > 自由回答 +- **底线**：不确定如实说，不猜不编 +- **品牌保护**：传销问题必须正面肯定回答是合法企业 + +**实际下发**：`antiThinkingPrefix` + 空格 + `DEFAULT_VOICE_SYSTEM_ROLE`，拼接后通过 `normalizeTextForSpeech()` 清洗 + +--- + +### 1.3 说话风格（DEFAULT_VOICE_SPEAKING_STYLE） + +**位置**：line 35，常量定义 + +``` +整体语气亲切自然、轻快有温度，像熟悉行业的朋友在语音聊天。优先短句和口语化表达，先给结论，再补一句最有帮助的信息。不要播音腔，不要念稿，不要客服腔，不要过度热情，也不要输出任何思考过程。 +``` + +**关键设计点**： +- 五个"不要"：播音腔、念稿、客服腔、过度热情、思考过程 +- 结构：先结论，再补充 +- 下发字段：S2S payload 的 `dialog.speaking_style` + +--- + +### 1.4 开场白（DEFAULT_VOICE_GREETING） + +**位置**：line 37，常量定义 + +``` +嗨，你好呀，我是大沃。你想了解德国PM产品、健康营养，还是一成系统和合作这块，我都可以跟你聊。 +``` + +**设计目的**：语音通话接通后的第一句话，自然引导用户提问方向。 + +--- + +### 1.5 审核拦截回复（audit_response） + +**位置**：`buildStartSessionPayload()` → `dialog.extra.audit_response`，line 88 + +``` +抱歉，这个问题我暂时无法回答。 +``` + +**触发条件**：火山引擎侧内容审核触发时的替代回复。当前 `strict_audit: false`。 + +--- + +### 1.6 客户端可覆盖 + +以上人设支持客户端通过 WebSocket `start` 消息覆盖（line 967-973）： + +| 字段 | 客户端参数 | 默认值 | +|------|-----------|--------| +| 机器人名 | `botName` | `大沃` | +| 系统角色 | `systemRole` | `DEFAULT_VOICE_SYSTEM_ROLE` | +| 说话风格 | `speakingStyle` | `DEFAULT_VOICE_SPEAKING_STYLE` | +| 开场白 | `greetingText` | `DEFAULT_VOICE_GREETING` | +| 音色 | `speaker` | `zh_female_vv_jupiter_bigtts` | +| 模型版本 | `modelVersion` | `O` | + +--- + +## 2. 知识库检索层（方舟 API） + +> 文件：`toolExecutor.js` +> 作用：通过方舟 Chat Completions API + knowledge_base metadata 进行知识检索 + +### 2.1 知识库回答模式（baseAnswerPrompt） + +**位置**：`searchArkKnowledge()` 内，line 911 + +``` +你是大沃，德国PM-FitLine产品专家。你的回答必须严格依据知识库内容，不得补充知识库未提及的信息，不得猜测，不得编造。若知识库中没有明确答案，就直接说明知识库未提及或暂未找到相关信息。回答保持口语化、简洁、专业，200字内。 +``` + +**关键设计点**： +- 人设一致：与 S2S 层的"大沃"身份统一 +- **严格知识库策略**：只能依据知识库回答，禁止补充、猜测、编造 +- **无答案兜底**：知识库没有明确答案时，必须直接说明未提及或未找到 +- 字数限制：200字内 +- 配合参数：`max_tokens=80`, `thinking: { type: 'disabled' }` + +--- + +### 2.2 知识库片段提取模式（baseSnippetPrompt） + +**位置**：`searchArkKnowledge()` 内，line 910 + +``` +知识库片段提取助手。提取2-4条与问题最相关的简洁事实片段。只输出中文事实，不寒暄，不写"根据知识库"，不补充未出现的内容，无相关内容则说未找到。 +``` + +**触发条件**：`responseMode === 'snippet'` 时使用，用于预查询等场景。 + +--- + +### 2.3 聚焦指令（动态拼接） + +**位置**：line 915-916，当 `responseMode === 'answer'` 时动态拼接 + +``` +当前必须优先直接回答用户当前这一个问题："${answerTargetQuery}"。如果用户只问一个维度，例如成分、价格、用法、适合谁、区别、正规性、地址或联系方式，就只回答这个维度，不要扩展成整段产品或公司介绍。 +``` + +**设计目的**：防止 LLM 收到知识库文档后"展开长篇大论"，强制聚焦用户的具体问题维度。 + +--- + +### 2.4 问题维度槽位指令（buildQuestionSlotInstruction） + +**位置**：line 200-218，`buildQuestionSlotInstruction()` 静态方法 + +根据用户问题自动分类到 14 个维度，每个维度有专属约束指令： + +| 维度 | 槽位 | 指令内容 | +|------|------|---------| +| 价格 | `price` | 用户当前只关心价格或费用，请只回答价格、收费或是否未提及价格，不要扩展到产品总介绍。 | +| 成分 | `ingredient` | 用户当前只关心成分或配方，请只回答成分、原料或是否未提及成分，不要扩展到品牌背景。 | +| 用法 | `usage` | 用户当前只关心用法、吃法、服用频次或剂量，请只回答这一点。 | +| 副作用 | `side_effect` | 用户当前只关心副作用或好转反应，请只回答可能的不良反应、好转反应或注意事项。 | +| 见效时间 | `effect_time` | 用户当前只关心多久见效或效果周期，请只回答见效时间、周期或个体差异，不要扩展无关信息。 | +| 医疗声明 | `medical_claim` | 用户当前只关心产品能不能治病、是不是药，请只回答是否属于药品、能否替代药物以及相关注意事项。 | +| 搭配原因 | `bundle_reason` | 用户当前只关心为什么要全套、搭配或三合一，请只回答搭配原理、协同作用或NTC相关原因。 | +| 事业发展 | `business_growth` | 用户当前只关心PM事业发展、线上拓客、陌生客户沟通、一成系统赋能、三大平台四大Ai生态或自我介绍，请只回答这类业务发展问题。 | +| 功效 | `benefit` | 用户当前只关心功效或作用，请只回答作用点，不要扩展到无关信息。 | +| 适用人群 | `audience` | 用户当前只关心适合人群，请只回答适用对象。 | +| 合法性 | `legality` | 用户当前只关心正规性、合法性或是否传销，请只围绕合法合规问题直接回答。 | +| 地址联系 | `address_contact` | 用户当前只关心地址或联系方式，请只回答地址、电话、联系信息。 | +| 对比差异 | `difference` | 用户当前只关心区别或对比，请直接做差异对比，不要扩写成单个产品长介绍。 | +| 通用 | `general` | 请优先直接回答用户当前这一问，不要离题扩展。 | + +--- + +### 2.5 改写注入指令（hasRewrite） + +**位置**：line 918-920，当检索词≠原始问题时动态拼接 + +``` +重要：用户的实际问题是"${cleanOriginal}"，请围绕这个问题回答，不要偏离用户的真实意图。下方的检索词仅用于匹配知识库文档，不代表用户的真正提问。 +``` + +**触发条件**：`rewriteKnowledgeQuery()` 改写了查询词（如 "怎么吃" → "德国PM细胞营养素 基础套装 大白 小红 小白"），LLM 需要知道用户实际问的是什么。 + +--- + +### 2.6 未命中回复模板 + +**位置**：`classifyKnowledgeAnswer()` 内，line 572/581/599/877 + +``` +知识库中暂未找到与"${query}"直接相关的信息，请换个更具体的问法再试。 +``` + +--- + +## 3. 对话交接层（Handoff） + +> 文件：`arkChatService.js` + `realtimeDialogRouting.js` +> 作用：文字聊天→语音通话切换时，生成对话摘要供语音模型接管 + +### 3.1 交接摘要生成指令 + +**位置**：`arkChatService.js` → `summarizeContextForHandoff()`，line 76 + +``` +你是对话交接摘要助手。请基于最近几轮对话生成一段简洁中文摘要，供另一个模型无缝接管会话。摘要必须同时包含：用户当前主要问题、已经确认的信息、仍待解决的问题。不要使用标题、项目符号或编号，不要虚构事实，控制在120字以内。 +``` + +**设计目的**：当用户从文字聊天切换到语音通话时，压缩历史对话为 120 字摘要，注入到语音会话上下文头部。 + +### 3.2 摘要注入格式 + +**位置**：`realtimeDialogRouting.js` → `withHandoffSummary()`，line 159 + +``` +会话交接摘要：${summary} +``` + +以 `role: 'assistant'` 消息形式注入到上下文最前面，一次性使用后标记 `handoffSummaryUsed=true`。 + +--- + +## 4. 内容安全层（SafeGuard） + +> 文件：`contentSafeGuard.js` + `realtimeDialogRouting.js` +> 作用：拦截 AI 输出中的品牌有害内容，替换为安全回复 + +### 4.1 语音模式安全回复（VOICE_SAFE_REPLY） + +**位置**：`contentSafeGuard.js`，line 119 + +``` +不好意思，我刚才没有听清楚，你可以再说一遍吗？ +``` + +**策略**：语音场景中，假装没听清重新引导，避免生硬拦截。 + +### 4.2 文字模式安全回复（TEXT_SAFE_REPLY） + +**位置**：`contentSafeGuard.js`，line 122 + +``` +德国PM是一家1993年成立于德国的合法直销公司，获得邓白氏AAA+认证，业务覆盖100多个国家和地区。如果你想了解更多，可以问我关于PM公司的详细介绍哦。 +``` + +**策略**：文字场景中，正面回应品牌合法性。 + +### 4.3 品牌保护回复（safeReply — 传销/正规性） + +**位置**：`realtimeDialogRouting.js` → `resolveReply()`，line 397 + +``` +德国PM是一家1993年成立于德国的合法直销公司，获得邓白氏AAA+认证，业务覆盖100多个国家和地区。它不是传销，是正规的直销企业哦。如果你想了解更多，可以问我关于PM公司或产品的详细介绍。 +``` + +**触发条件**：用户问传销/正规性相关问题，知识库未命中时（`!toolResult.hit`），直接返回此预设回复，不交给 S2S 自由发挥。 + +### 4.4 KB保护窗口诚实兜底（honestReply） + +**位置**：`realtimeDialogRouting.js`，line 411 + +``` +这个问题我暂时不太确定具体细节，建议你咨询一下你的推荐人，或者换个更具体的问法再问我。 +``` + +**触发条件**：KB 最近 60 秒内命中过（保护窗口），新问题 no-hit 时使用此回复，防止 S2S 自由编造产品信息。 + +### 4.5 内容安全检测策略（三层） + +| 层级 | 策略 | 示例 | +|------|------|------| +| 第一层 | 通用负面词（58个），单独出现即拦截 | 传销、骗局、智商税、洗脑、割韭菜 | +| 第二层 | 品牌名 + 负面后缀组合（28个） | PM是传销、FitLine不靠谱 | +| 第三层 | 负面前缀 + 品牌名组合（14个） | 骗子公司PM、远离FitLine | +| 白名单 | 正面合法性描述放行（15个） | 不是传销、合法直销企业、邓白氏AAA+ | + +--- + +## 5. 高频热答案（HOT_ANSWERS） + +> 文件：`toolExecutor.js`，line 75-160 +> 作用：本地维护的高频问答内容资产，当前保留在代码中，但**不再在 `answer` 模式下直接返回给用户**，以避免绕过知识库。 + +共 **21 条**预设问答，覆盖最高频场景： + +| # | 话题 | 匹配模式示例 | 答案摘要 | +|---|------|-------------|---------| +| 1 | 基础三合一吃法 | `基础三合一.*怎么吃` | 大白空腹→小红15-30分钟后→小白睡前 | +| 2 | 传销/正规性 | `PM.*传销`, `是不是传销` | 1993年成立，邓白氏AAA+，100+国家 | +| 3 | NTC核心优势 | `NTC.*核心优势`, `营养保送系统` | 精准保送到细胞层面 | +| 4 | 见效时间 | `多久见效`, `什么时候见效` | 小红当天感受→整体1-3个月 | +| 5 | 全套搭配原因 | `为什么.*全套`, `为什么.*三合一` | NTC协同+火炉原理 | +| 6 | 好转反应 | `好转反应`, `副作用` | 正常整应反应，3-7天消失 | +| 7 | 公司介绍 | `PM.*公司介绍`, `德国PM介绍` | 1993年，FitLine，NTC，邓白氏AAA+ | +| 8 | 小红功效 | `小红.*功效`, `Activize.*作用` | 提升细胞能量，B族+C+Q10 | +| 9 | 大白功效 | `大白.*功效`, `Basics.*作用` | 基础营养素，维生素矿物质 | +| 10 | 小白功效 | `小白.*功效`, `Restorate.*作用` | 夜间修复，矿物质微量元素 | +| 11 | 保健品区别 | `保健品.*区别` | NTC保送vs普通补充 | +| 12 | CC套装 | `CC.*套装`, `CC-Cell` | 胶囊+乳霜，抗衰产品 | +| 13 | Q10 | `Q10.*功效`, `辅酵素` | 抗氧化，心脏，皮肤弹性 | +| 14 | IB5 | `IB5`, `口腔.*喷雾` | 口腔免疫喷雾，益生菌 | +| 15 | 邓白氏认证 | `邓白氏.*认证`, `AAA+` | 全球最权威商业信用评估 | +| 16 | 一成系统 | `一成系统.*是什么`, `三大平台.*四大` | 三大平台+四大AI生态 | +| 17 | 火炉原理 | `火炉原理`, `暖炉原理` | 柴火+引火物+氧气比喻 | +| 18 | D-Drink | `D-Drink`, `14天排毒` | 14天排毒饮料 | +| 19 | 加入方式 | `如何加入`, `怎么做PM` | 联系推荐人注册 | +| 20 | 特殊人群 | `孕妇.*能吃`, `儿童.*能吃` | 建议咨询医生 | +| 21 | 价格/购买 | `多少钱`, `怎么买` | 咨询推荐人，会员优惠价 | + +--- + +## 6. ASR 热词上下文 + +> 文件：`nativeVoiceGateway.js` → `buildStartSessionPayload()` → `asr.extra.context` +> 作用：提升火山 ASR 对专业术语的识别准确率 + +**位置**：line 67 + +``` +一成,一成系统,大沃,PM,PM-FitLine,FitLine,细胞营养素,Ai众享,AI众享,盛咖学愿,数字化工作室, +Activize,Basics,Restorate,NTC,基础三合一,招商,阿育吠陀,小红产品,小红,小白,大白,肽美,艾特维, +德丽,德维,宝丽,美固健,Activize Oxyplus,Basic Power,CitrusCare,NutriSunny,Q10,Omega, +葡萄籽,白藜芦醇,益生菌,胶原蛋白肽,Germany,FitLine细胞营养,FitLine营养素,德国PM营养素, +德国PM FitLine,德国PM细胞营养,德国PM产品,德国PM健康,德国PM事业,德国PM招商,一成,一成团队, +一成商学院,数字化,数字化运营,数字化经营,数字化营销,数字化创业,数字化工作室,数字化事业, +招商加盟,合作加盟,事业合作 +``` + +--- + +## 附：提示词流转关系图 + +``` +用户语音输入 + │ + ▼ +┌─────────────────────────────┐ +│ 火山 S2S（语音对话模型） │ +│ │ +│ system_role: │ +│ antiThinkingPrefix │ +│ + DEFAULT_VOICE_SYSTEM_ROLE │ +│ │ +│ speaking_style: │ +│ DEFAULT_VOICE_SPEAKING_STYLE │ +│ │ +│ greeting: │ +│ DEFAULT_VOICE_GREETING │ +└──────────┬──────────────────┘ + │ 路由到知识库 + ▼ +┌─────────────────────────────┐ +│ 方舟 KB API（知识库检索） │ +│ │ +│ system: │ +│ baseAnswerPrompt │ +│ + 聚焦指令 │ +│ + 槽位指令 │ +│ + 改写注入指令（可选） │ +└──────────┬──────────────────┘ + │ 回复内容 + ▼ +┌─────────────────────────────┐ +│ 内容安全检测 │ +│ │ +│ 三层有害内容检测 │ +│ ↓ 有害 → VOICE_SAFE_REPLY │ +│ ↓ 传销no-hit → safeReply │ +│ ↓ 保护窗口no-hit → honestReply │ +│ ↓ 正常 → 原文输出 │ +└─────────────────────────────┘ +``` + +--- + +## 附：环境变量对提示词的影响 + +| 环境变量 | 影响 | 当前值 | +|---------|------|--------| +| `VOLC_ARK_KB_MODEL` | KB检索使用的模型 | `ep-20260320175538-lcg7g` (Seed-2.0-lite) | +| `VOLC_ARK_KNOWLEDGE_THRESHOLD` | KB检索相似度阈值 | `0.3` | +| `VOLC_ARK_KNOWLEDGE_TOP_K` | KB检索返回条数 | `3` | +| `VOLC_S2S_SPEAKER_ID` | TTS音色 | `zh_female_vv_jupiter_bigtts` | + diff --git a/test2/server/routes/assistantProfile.js b/test2/server/routes/assistantProfile.js new file mode 100644 index 0000000..916ad13 --- /dev/null +++ b/test2/server/routes/assistantProfile.js @@ -0,0 +1,65 @@ +const express = require('express'); +const router = express.Router(); +const { getAssistantProfile, clearAssistantProfileCache } = require('../services/assistantProfileService'); + +router.get('/', async (req, res) => { + try { + const userId = String(req.query.userId || '').trim() || null; + const forceRefresh = String(req.query.forceRefresh || '').trim() === 'true'; + const result = await getAssistantProfile({ userId, forceRefresh }); + res.json({ + code: 0, + message: 'success', + success: true, + data: { + userId, + profile: result.profile, + source: result.source, + cached: result.cached, + fetchedAt: result.fetchedAt, + configured: result.configured, + error: result.error, + }, + }); + } catch (error) { + console.error('[AssistantProfile] query failed:', error.message); + res.status(500).json({ + code: 500, + message: error.message, + success: false, + error: error.message, + }); + } +}); + +router.post('/refresh', async (req, res) => { + try { + const userId = String(req.body?.userId || '').trim() || null; + clearAssistantProfileCache(userId); + const result = await getAssistantProfile({ userId, forceRefresh: true }); + res.json({ + code: 0, + message: 'success', + success: true, + data: { + userId, + profile: result.profile, + source: result.source, + cached: result.cached, + fetchedAt: result.fetchedAt, + configured: result.configured, + error: result.error, + }, + }); + } catch (error) { + console.error('[AssistantProfile] refresh failed:', error.message); + res.status(500).json({ + code: 500, + message: error.message, + success: false, + error: error.message, + }); + } +}); + +module.exports = router; diff --git a/test2/server/routes/chat.js b/test2/server/routes/chat.js index 76aa073..770618f 100644 --- a/test2/server/routes/chat.js +++ b/test2/server/routes/chat.js @@ -3,14 +3,14 @@ const router = express.Router(); const cozeChatService = require('../services/cozeChatService'); const arkChatService = require('../services/arkChatService'); const ToolExecutor = require('../services/toolExecutor'); +const contextKeywordTracker = require('../services/contextKeywordTracker'); const { shouldForceKnowledgeRoute } = require('../services/realtimeDialogRouting'); +const { isBrandHarmful, getTextSafeReply } = require('../services/contentSafeGuard'); const db = require('../db'); // 存储文字对话的会话状态（sessionId -> session） const chatSessions = new Map(); -const BRAND_HARMFUL_PATTERN = /传销|骗局|骗子公司|非法集资|非法经营|不正规|不合法|庞氏骗局|老鼠会|拉人头的|割韭菜/; -const BRAND_SAFE_REPLY = '德国PM是一家1993年成立于德国的合法直销公司，获得邓白氏AAA+认证，业务覆盖100多个国家和地区。如果你想了解更多，可以问我关于PM公司的详细介绍哦。'; function normalizeAssistantText(text) { let result = String(text || '') @@ -22,9 +22,9 @@ function normalizeAssistantText(text) { .replace(/([。！？；，])\s*([。！？；，])/g, '$2') .replace(/\s+/g, ' ') .trim(); - if (BRAND_HARMFUL_PATTERN.test(result)) { + if (isBrandHarmful(result)) { console.warn(`[Chat][SafeGuard] blocked harmful content: ${JSON.stringify(result.slice(0, 200))}`); - return BRAND_SAFE_REPLY; + return getTextSafeReply(); } return result; } @@ -52,16 +52,46 @@ async function loadHandoffMessages(sessionId, voiceSubtitles = []) { return voiceMessages; } -async function buildChatSessionState(sessionId, voiceSubtitles = []) { - const voiceMessages = await loadHandoffMessages(sessionId, voiceSubtitles); - let handoffSummary = ''; - try { - handoffSummary = await arkChatService.summarizeContextForHandoff(voiceMessages, 3); - } catch (error) { - console.warn('[Chat] summarizeContextForHandoff failed:', error.message); +function buildDeterministicHandoffSummary(messages = []) { + const normalizedMessages = (Array.isArray(messages) ? messages : []) + .filter((item) => item && (item.role === 'user' || item.role === 'assistant') && String(item.content || '').trim()) + .slice(-8); + if (!normalizedMessages.length) { + return ''; } + const userMessages = normalizedMessages.filter((item) => item.role === 'user'); + const currentQuestion = String(userMessages[userMessages.length - 1]?.content || '').trim(); + const previousQuestion = String(userMessages[userMessages.length - 2]?.content || '').trim(); + const assistantFacts = normalizedMessages + .filter((item) => item.role === 'assistant') + .slice(-2) + .map((item) => String(item.content || '').trim()) + .filter(Boolean) + .map((item) => item.slice(0, 60)) + .join('；'); + const parts = []; + if (currentQuestion) { + parts.push(`当前问题：${currentQuestion}`); + } + if (previousQuestion && previousQuestion !== currentQuestion) { + parts.push(`上一轮关注：${previousQuestion}`); + } + if (assistantFacts) { + parts.push(`已给信息：${assistantFacts}`); + } + return parts.join('；'); +} + +async function buildChatSessionState(sessionId, voiceSubtitles = [], userId = null) { + const voiceMessages = await loadHandoffMessages(sessionId, voiceSubtitles); + voiceMessages + .filter((item) => item.role === 'user') + .slice(-6) + .forEach((item) => contextKeywordTracker.updateSession(sessionId, item.content)); + const handoffSummary = buildDeterministicHandoffSummary(voiceMessages); return { - userId: `user_${sessionId.slice(0, 12)}`, + userId: userId || `user_${sessionId.slice(0, 12)}`, + profileUserId: userId || null, conversationId: null, voiceMessages, handoffSummary, @@ -127,7 +157,7 @@ async function tryKnowledgeReply(sessionId, session, message) { if (!shouldForceKnowledgeRoute(text, context)) { return null; } - const result = await ToolExecutor.execute('search_knowledge', { query: text }, context); + const result = await ToolExecutor.execute('search_knowledge', { query: text, session_id: sessionId, _session: { userId: session?.userId || null, profileUserId: session?.profileUserId || null } }, context); if (!result?.hit) { return null; } @@ -160,7 +190,7 @@ async function tryKnowledgeReply(sessionId, session, message) { * 创建文字对话会话，可选传入语音通话的历史字幕 */ router.post('/start', async (req, res) => { - const { sessionId, voiceSubtitles = [] } = req.body; + const { sessionId, voiceSubtitles = [], userId = null } = req.body; if (!sessionId) { return res.status(400).json({ success: false, error: 'sessionId is required' }); @@ -170,10 +200,10 @@ router.post('/start', async (req, res) => { return res.status(500).json({ success: false, error: 'Coze 智能体未配置，请设置 COZE_API_TOKEN 和 COZE_BOT_ID' }); } - const sessionState = await buildChatSessionState(sessionId, voiceSubtitles); + const sessionState = await buildChatSessionState(sessionId, voiceSubtitles, userId); // 更新数据库会话模式为 chat - try { await db.createSession(sessionId, `user_${sessionId.slice(0, 12)}`, 'chat'); } catch (e) {} + try { await db.createSession(sessionId, userId || `user_${sessionId.slice(0, 12)}`, 'chat'); } catch (e) {} chatSessions.set(sessionId, sessionState); @@ -195,7 +225,7 @@ router.post('/start', async (req, res) => { */ router.post('/send', async (req, res) => { try { - const { sessionId, message } = req.body; + const { sessionId, message, userId = null } = req.body; if (!sessionId || !message) { return res.status(400).json({ success: false, error: 'sessionId and message are required' }); @@ -205,10 +235,15 @@ router.post('/send', async (req, res) => { // 自动创建会话（如果不存在） if (!session) { - session = await buildChatSessionState(sessionId, []); + session = await buildChatSessionState(sessionId, [], userId); chatSessions.set(sessionId, session); } + if (userId) { + session.userId = userId; + session.profileUserId = userId; + } session.lastActiveAt = Date.now(); + contextKeywordTracker.updateSession(sessionId, message); console.log(`[Chat] User(${sessionId}): ${message}`); @@ -296,7 +331,7 @@ router.get('/history/:sessionId', (req, res) => { * 流式发送文字消息（SSE），逐字输出 Coze 智能体回复 */ router.post('/send-stream', async (req, res) => { - const { sessionId, message } = req.body; + const { sessionId, message, userId = null } = req.body; if (!sessionId || !message) { return res.status(400).json({ success: false, error: 'sessionId and message are required' }); @@ -304,10 +339,15 @@ router.post('/send-stream', async (req, res) => { let session = chatSessions.get(sessionId); if (!session) { - session = await buildChatSessionState(sessionId, []); + session = await buildChatSessionState(sessionId, [], userId); chatSessions.set(sessionId, session); } + if (userId) { + session.userId = userId; + session.profileUserId = userId; + } session.lastActiveAt = Date.now(); + contextKeywordTracker.updateSession(sessionId, message); console.log(`[Chat][SSE] User(${sessionId}): ${message}`); @@ -339,6 +379,10 @@ router.post('/send-stream', async (req, res) => { // 首次对话时注入语音历史作为上下文 const extraMessages = !session.conversationId ? buildInitialContextMessages(session) : []; + // 流式缓冲检测：累积 chunk 内容，实时检测有害关键词 + let streamBuffer = ''; + let harmfulDetected = false; + const result = await cozeChatService.chatStream( session.userId, message, @@ -346,24 +390,36 @@ router.post('/send-stream', async (req, res) => { extraMessages, { onChunk: (text) => { + if (harmfulDetected) return; + streamBuffer += text; + // 实时检测流式内容是否包含有害关键词 + if (isBrandHarmful(streamBuffer)) { + harmfulDetected = true; + console.warn(`[Chat][SSE][SafeGuard] harmful content detected in stream, intercepting session=${sessionId} buffer=${JSON.stringify(streamBuffer.slice(0, 200))}`); + // 发送重置信号，告诉前端丢弃已收到的 chunk + res.write(`data: ${JSON.stringify({ type: 'stream_reset', reason: 'content_safety' })}\n\n`); + return; + } res.write(`data: ${JSON.stringify({ type: 'chunk', content: text })}\n\n`); }, onDone: () => {}, } ); - const normalizedContent = normalizeAssistantText(result.content); + + // 如果流式过程中检测到有害内容，使用安全回复替换 + const finalContent = harmfulDetected ? getTextSafeReply() : normalizeAssistantText(result.content); // 保存 Coze 返回的 conversationId session.conversationId = result.conversationId; session.handoffSummaryUsed = true; - console.log(`[Chat][SSE] Assistant(${sessionId}): ${normalizedContent?.substring(0, 100)}`); + console.log(`[Chat][SSE] Assistant(${sessionId}): ${finalContent?.substring(0, 100)}${harmfulDetected ? ' [SAFE_REPLACED]' : ''}`); // 写入数据库：AI 回复 - if (normalizedContent) { - db.addMessage(sessionId, 'assistant', normalizedContent, 'chat_bot').catch(e => console.warn('[DB] addMessage failed:', e.message)); + if (finalContent) { + db.addMessage(sessionId, 'assistant', finalContent, 'chat_bot').catch(e => console.warn('[DB] addMessage failed:', e.message)); } - res.write(`data: ${JSON.stringify({ type: 'done', content: normalizedContent })}\n\n`); + res.write(`data: ${JSON.stringify({ type: 'done', content: finalContent })}\n\n`); res.end(); } catch (error) { console.error('[Chat][SSE] Stream failed:', error.message); diff --git a/test2/server/routes/voice.js b/test2/server/routes/voice.js index 36850ba..521f62d 100644 --- a/test2/server/routes/voice.js +++ b/test2/server/routes/voice.js @@ -2,6 +2,8 @@ const express = require('express'); const router = express.Router(); const { v4: uuidv4 } = require('uuid'); const ToolExecutor = require('../services/toolExecutor'); +const contextKeywordTracker = require('../services/contextKeywordTracker'); +const { getRuleBasedDirectRouteDecision, shouldForceKnowledgeRoute } = require('../services/realtimeDialogRouting'); const DEFAULT_TOOLS = require('../config/tools'); const db = require('../db'); @@ -66,6 +68,9 @@ router.post('/direct/message', async (req, res) => { if (!sessionId || !text || !source) { return res.status(400).json({ success: false, error: 'sessionId, text and source are required' }); } + if (role === 'user') { + contextKeywordTracker.updateSession(sessionId, text); + } await db.addMessage(sessionId, role === 'user' ? 'user' : 'assistant', text, source, toolName || null); res.json({ success: true }); } catch (error) { @@ -94,9 +99,19 @@ router.post('/direct/query', async (req, res) => { const context = await db.getHistoryForLLM(sessionId, 20).catch(() => []); const cleanQuery = (query || '').trim(); if (appendUserMessage && cleanQuery) { + contextKeywordTracker.updateSession(sessionId, cleanQuery); await db.addMessage(sessionId, 'user', cleanQuery, 'voice_asr').catch(() => null); } - const result = await ToolExecutor.execute('search_knowledge', { query: cleanQuery }, context); + if (!appendUserMessage && cleanQuery) { + contextKeywordTracker.updateSession(sessionId, cleanQuery); + } + const routeDecision = getRuleBasedDirectRouteDecision(cleanQuery); + const forceKb = shouldForceKnowledgeRoute(cleanQuery, context); + const shouldSearchKb = routeDecision.route === 'search_knowledge' || forceKb; + const directSession = directSessions.get(sessionId); + const result = shouldSearchKb + ? await ToolExecutor.execute('search_knowledge', { query: cleanQuery, session_id: sessionId, _session: { userId: directSession?.userId || null } }, context) + : { hit: false, reason: 'route_skip', source: 'route_skip', error: '该问题不在知识库范围内，请咨询其他问题。' }; let contentText = JSON.stringify(result); if (result && result.results && Array.isArray(result.results)) { contentText = result.results.map((item) => item.content || JSON.stringify(item)).join('\n'); diff --git a/test2/server/services/arkChatService.js b/test2/server/services/arkChatService.js index 71e13f5..8620190 100644 --- a/test2/server/services/arkChatService.js +++ b/test2/server/services/arkChatService.js @@ -30,7 +30,7 @@ class ArkChatService { if (datasetIds.length === 0) return null; const topK = parseInt(process.env.VOLC_ARK_KNOWLEDGE_TOP_K) || 3; - const threshold = parseFloat(process.env.VOLC_ARK_KNOWLEDGE_THRESHOLD) || 0.5; + const threshold = parseFloat(process.env.VOLC_ARK_KNOWLEDGE_THRESHOLD) || 0.4; return { dataset_ids: datasetIds, diff --git a/test2/server/services/assistantProfileConfig.js b/test2/server/services/assistantProfileConfig.js index 91956bb..a4872ec 100644 --- a/test2/server/services/assistantProfileConfig.js +++ b/test2/server/services/assistantProfileConfig.js @@ -73,7 +73,7 @@ function buildKnowledgeAnswerPrompt(profileOverrides = null) { const personalInfoBlock = personalInfoLines.length > 0 ? ` 对于${profile.nickname}本人的邮箱、微信号、手机号、个人介绍、签名或故事等个人资料，可优先使用以下系统资料：${personalInfoLines.join(' ')}` : ''; - return `你是${profile.nickname}的智能助手${documentsClause}。你的回答必须严格依据知识库内容，不得补充知识库未提及的信息，不得猜测，不得编造。若知识库中没有明确答案，就直接说明知识库未提及或暂未找到相关信息。回答保持口语化、简洁、专业，200字内。${personalInfoBlock}`; + return `你是${profile.nickname}的智能助手${documentsClause}。知识库涵盖近50款PM-FitLine产品的完整资料（成分、用法、剂量、价格、规格、搭配方案、好转反应等）及117个常见问答。回答规则：产品相关具体信息必须严格依据知识库，不得猜测或自行补充；公司背景、健康常识可适当补充。产品常有别名（小红=艾特维、大白=倍适、小白=维适多等），请注意识别。不得编造产品名或数据。PM是营养品非药物，涉及疾病建议咨询医生。若知识库无相关内容，坦诚说明并建议咨询推荐人。回答口语化、简洁，1-3句给结论，150字内。${personalInfoBlock}`; } module.exports = { diff --git a/test2/server/services/assistantProfileService.js b/test2/server/services/assistantProfileService.js new file mode 100644 index 0000000..77fa205 --- /dev/null +++ b/test2/server/services/assistantProfileService.js @@ -0,0 +1,178 @@ +const axios = require('axios'); +const { + DEFAULT_VOICE_ASSISTANT_PROFILE, + resolveAssistantProfile, +} = require('./assistantProfileConfig'); + +const assistantProfileCache = new Map(); + +function getAssistantProfileApiUrl() { + const value = String(process.env.ASSISTANT_PROFILE_API_URL || '').trim(); + return value && !value.startsWith('your_') ? value : ''; +} + +function getAssistantProfileApiMethod() { + const method = String(process.env.ASSISTANT_PROFILE_API_METHOD || 'GET').trim().toUpperCase(); + return method === 'POST' ? 'POST' : 'GET'; +} + +function getAssistantProfileTimeoutMs() { + const value = Number(process.env.ASSISTANT_PROFILE_API_TIMEOUT_MS || 5000); + return Number.isFinite(value) && value > 0 ? value : 5000; +} + +function getAssistantProfileCacheTtlMs() { + const value = Number(process.env.ASSISTANT_PROFILE_CACHE_TTL_MS || 60000); + return Number.isFinite(value) && value >= 0 ? value : 60000; +} + +function getAssistantProfileCacheKey(userId = null) { + return String(userId || 'global').trim() || 'global'; +} + +function parseAssistantProfileHeaders() { + const raw = String(process.env.ASSISTANT_PROFILE_API_HEADERS || '').trim(); + if (!raw) return {}; + try { + const parsed = JSON.parse(raw); + if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) { + return {}; + } + return Object.fromEntries( + Object.entries(parsed) + .map(([key, value]) => [String(key || '').trim(), String(value || '').trim()]) + .filter(([key, value]) => key && value) + ); + } catch (error) { + console.warn('[AssistantProfile] parse headers failed:', error.message); + return {}; + } +} + +function pickAssistantProfilePayload(payload) { + if (!payload || typeof payload !== 'object' || Array.isArray(payload)) { + return {}; + } + const candidates = [ + payload.assistantProfile, + payload.profile, + payload.data?.assistantProfile, + payload.data?.profile, + payload.data, + payload, + ]; + for (const candidate of candidates) { + if (candidate && typeof candidate === 'object' && !Array.isArray(candidate)) { + return candidate; + } + } + return {}; +} + +function sanitizeAssistantProfilePayload(payload) { + const source = pickAssistantProfilePayload(payload); + return { + documents: source.documents, + email: source.email, + nickname: source.nickname, + wxl: source.wxl, + mobile: source.mobile, + wx_code: source.wx_code, + intro: source.intro, + sign: source.sign, + story: source.story, + }; +} + +async function fetchRemoteAssistantProfile(userId = null) { + const url = getAssistantProfileApiUrl(); + if (!url) { + return { + profile: resolveAssistantProfile(), + source: 'default', + cached: false, + fetchedAt: Date.now(), + configured: false, + error: null, + }; + } + const headers = { + Accept: 'application/json', + ...parseAssistantProfileHeaders(), + }; + const token = String(process.env.ASSISTANT_PROFILE_API_TOKEN || '').trim(); + if (token) { + headers.Authorization = `Bearer ${token}`; + } + const method = getAssistantProfileApiMethod(); + const timeout = getAssistantProfileTimeoutMs(); + const params = userId ? { userId } : undefined; + const response = method === 'POST' + ? await axios.post(url, userId ? { userId } : {}, { headers, timeout }) + : await axios.get(url, { headers, timeout, params }); + const profile = resolveAssistantProfile(sanitizeAssistantProfilePayload(response.data)); + return { + profile, + source: 'remote_api', + cached: false, + fetchedAt: Date.now(), + configured: true, + error: null, + }; +} + +async function getAssistantProfile(options = {}) { + const userId = String(options.userId || '').trim() || null; + const forceRefresh = !!options.forceRefresh; + const overrides = options.overrides && typeof options.overrides === 'object' ? options.overrides : null; + const cacheKey = getAssistantProfileCacheKey(userId); + const ttlMs = getAssistantProfileCacheTtlMs(); + const cached = assistantProfileCache.get(cacheKey); + + if (!forceRefresh && cached && (Date.now() - cached.fetchedAt) <= ttlMs) { + return { + profile: resolveAssistantProfile({ ...cached.profile, ...(overrides || {}) }), + source: cached.source, + cached: true, + fetchedAt: cached.fetchedAt, + configured: cached.configured, + error: null, + }; + } + + try { + const fetched = await fetchRemoteAssistantProfile(userId); + assistantProfileCache.set(cacheKey, fetched); + return { + profile: resolveAssistantProfile({ ...fetched.profile, ...(overrides || {}) }), + source: fetched.source, + cached: false, + fetchedAt: fetched.fetchedAt, + configured: fetched.configured, + error: null, + }; + } catch (error) { + const fallback = cached?.profile || DEFAULT_VOICE_ASSISTANT_PROFILE; + return { + profile: resolveAssistantProfile({ ...fallback, ...(overrides || {}) }), + source: cached ? 'cache_fallback' : 'default_fallback', + cached: !!cached, + fetchedAt: cached?.fetchedAt || null, + configured: !!getAssistantProfileApiUrl(), + error: error.message, + }; + } +} + +function clearAssistantProfileCache(userId = null) { + if (userId == null || String(userId).trim() === '') { + assistantProfileCache.clear(); + return; + } + assistantProfileCache.delete(getAssistantProfileCacheKey(userId)); +} + +module.exports = { + getAssistantProfile, + clearAssistantProfileCache, +}; diff --git a/test2/server/services/contentSafeGuard.js b/test2/server/services/contentSafeGuard.js new file mode 100644 index 0000000..ca3ec6f --- /dev/null +++ b/test2/server/services/contentSafeGuard.js @@ -0,0 +1,161 @@ +/** + * 内容安全兜底模块 + * 检测 AI 返回内容中的品牌有害关键词，拦截并替换为安全回复 + * + * 三层检测策略： + * 1. 通用负面词（不绑定品牌名，出现即拦截） + * 2. 品牌 + 负面组合词（品牌名 + 贬义描述） + * 3. 负面 + 品牌组合词（贬义描述 + 品牌名） + */ + +// ── 品牌名匹配片段 ── +const BRAND_NAMES = '德国PM|PM-International|PM公司|PM-FitLine|FitLine|一成系统|一成团队|大沃|PM营养素|PM健康|PM事业|PM直销|PM产品|PM'; + +// ── 第一层：通用负面关键词（单独出现即拦截） ── +const GENERIC_HARMFUL_WORDS = [ + // 传销 & 变体 + '传销', '直销骗局', '非法直销', '变相传销', '网络传销', '精神传销', + '传销组织', '传销模式', '传销公司', '传销骗局', '传销陷阱', '传销套路', + // 骗局 & 诈骗 + '骗局', '骗子公司', '骗子', '骗人的', '诈骗', '行骗', '欺诈', + '虚假宣传', '夸大宣传', '虚假广告', '消费欺诈', '商业欺诈', + // 非法 & 违法 + '非法集资', '非法经营', '非法营销', '非法组织', '非法敛财', + '涉嫌违法', '涉嫌传销', '疑似传销', '涉嫌欺诈', '涉嫌诈骗', + '违法经营', '违规经营', '违规操作', + // 不合规 + '不正规', '不合法', '不合规', '不靠谱', '不正当', + // 经典传销术语 + '庞氏骗局', '老鼠会', '拉人头', '割韭菜', '资金盘', '层级分销', + '金字塔骗局', '金字塔模式', '发展下线', '上线下线', + '会员费骗局', '入门费骗局', '人头费骗局', + // 社交媒体常见说法 + '智商税', '缴智商税', '交智商税', '收割', '被收割', '被割', + '洗脑', '被洗脑', '洗脑术', '洗脑营销', '精神控制', + '坑人', '坑钱', '坑货', '害人', '黑心', '黑幕', + '暴利', '暴利产品', '天价产品', '高价低质', + // 监管处罚 + '被查处', '被取缔', '被罚款', '被处罚', '被举报', + '工商处罚', '市场监管处罚', '行政处罚', + '依法处理', '依法查处', '依法取缔', + // 否定资质 + '没有合法直销资质', '没有直销资质', '不具备直销资质', + '没有合法资质', '没有经营资质', '无合法资质', + '没有取得批准文号', '未取得批准文号', '没取得批准文号', + '没有取得资质', '未取得资质', '没取得资质', + '没有取得直销资质', '未取得直销资质', + '没有取得牌照', '未取得牌照', + '没有保健食品批准', '未取得保健食品', + // 劝退 & 风险警告 + '法律风险', '资金损失', '经济损失', '血本无归', + '不符合相关法律', '不符合法律法规', '违反法律法规', + // 受害者视角 + '受害者', '受骗者', '上当受骗', '上当了', '被骗了', + '维权', '退款难', '投诉无门', + // 负面定性 + '臭名昭著', '声名狼藉', '劣迹斑斑', '口碑极差', + '过街老鼠', '千夫所指', +]; + +// ── 第二层：品牌名 + 负面描述组合 ── +const BRAND_NEGATIVE_SUFFIXES = [ + '是传销', '属于传销', '涉嫌传销', '疑似传销', '就是传销', + '是骗局', '是骗子', '是骗人的', '在骗人', '骗钱', + '是非法的', '不合法', '不正规', '不靠谱', '不可信', + '有问题', '有争议', '有风险', '有隐患', + '已被查', '已被处罚', '被取缔', '被举报', + '在割韭菜', '在收割', '在洗脑', '在坑人', '在骗人', + '没有资质', '没有牌照', '没有直销牌照', '没有合法直销资质', '没有直销资质', + '没取得', '没有取得', '未取得', + '不符合法律', '不符合法规', '不符合相关法律', + '存在法律风险', '带来法律风险', '面临法律风险', + '存在资金损失', '带来资金损失', '带来经济损失', + '产品不行', '产品是假的', '产品没用', '产品有害', + '害了很多人', '坑了很多人', '骗了很多人', + '不值得信任', '不值得加入', '不建议加入', '不要加入', + '千万别信', '千万别买', '千万不要', '别上当', + '是假的', '没用', '没效果', '没什么用', '不管用', '是垃圾', '垃圾产品', +]; + +// ── 第三层：负面描述 + 品牌名组合 ── +const NEGATIVE_BRAND_PREFIXES = [ + '传销组织', '传销公司', '传销骗局', '传销陷阱', + '骗局', '骗子公司', '骗人的', + '非法的', '不合法的', '不正规的', '不靠谱的', + '有争议的', '有问题的', '有风险的', + '坑人的', '害人的', '割韭菜的', '洗脑的', + '千万别信', '千万不要买', '千万不要加入', + '远离', '警惕', '小心', '当心', '注意', +]; + +// 构建正则 +const genericPattern = GENERIC_HARMFUL_WORDS.map((w) => w.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')).join('|'); +const brandSuffixPattern = `(?:${BRAND_NAMES}).*?(?:${BRAND_NEGATIVE_SUFFIXES.map((w) => w.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')).join('|')})`; +const prefixBrandPattern = `(?:${NEGATIVE_BRAND_PREFIXES.map((w) => w.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')).join('|')}).*?(?:${BRAND_NAMES})`; + +const BRAND_HARMFUL_PATTERN = new RegExp( + `${genericPattern}|${brandSuffixPattern}|${prefixBrandPattern}`, + 'i', +); + +// ── 白名单：正面合法性描述，匹配则放行 ── +const POSITIVE_PHRASES = [ + '不是传销', '并非传销', '并不是传销', '不属于传销', + '合法正规的直销企业', '合法正规直销企业', '合法直销公司', '合法直销企业', + '正规直销企业', '正规直销公司', '正规持牌直销公司', '正规持牌直销企业', + '拥有直销牌照', '持有直销牌照', '获得直销牌照', + '邓白氏AAA\\+', '邓白氏AAA', 'AAA\\+认证', 'AAA\\+信用', + '合法合规', '正规合法', '正规经营', + '业务覆盖全球', '覆盖.*国家', + '1993年成立', '成立于德国', +]; +const positivePhrasesPattern = POSITIVE_PHRASES.join('|'); +const BRAND_POSITIVE_LEGALITY_PATTERN = new RegExp( + `(?:${BRAND_NAMES}).*?(?:${positivePhrasesPattern})|(?:${positivePhrasesPattern}).*?(?:${BRAND_NAMES})`, + 'i', +); + +// 语音模式安全回复（假装没听清，让用户重新说） +const VOICE_SAFE_REPLY = '不好意思，我刚才没有听清楚，你可以再说一遍吗？'; + +// 文字模式安全回复（正面回应品牌合法性） +const TEXT_SAFE_REPLY = '德国PM是一家1993年成立于德国的合法直销公司，获得邓白氏AAA+认证，业务覆盖100多个国家和地区。如果你想了解更多，可以问我关于PM公司的详细介绍哦。'; + +/** + * 检测文本是否包含品牌有害内容 + * @param {string} text - 待检测文本 + * @returns {boolean} true = 有害，需要拦截 + */ +function isBrandHarmful(text) { + if (!text) return false; + const normalized = String(text).replace(/\s+/g, ' '); + // 白名单放行：正面描述 PM 合法性的内容 + if (BRAND_POSITIVE_LEGALITY_PATTERN.test(normalized)) { + return false; + } + return BRAND_HARMFUL_PATTERN.test(normalized); +} + +/** + * 获取语音模式的安全回复 + */ +function getVoiceSafeReply() { + return VOICE_SAFE_REPLY; +} + +/** + * 获取文字模式的安全回复 + */ +function getTextSafeReply() { + return TEXT_SAFE_REPLY; +} + +module.exports = { + BRAND_HARMFUL_PATTERN, + BRAND_POSITIVE_LEGALITY_PATTERN, + VOICE_SAFE_REPLY, + TEXT_SAFE_REPLY, + isBrandHarmful, + getVoiceSafeReply, + getTextSafeReply, +}; diff --git a/test2/server/services/contextKeywordTracker.js b/test2/server/services/contextKeywordTracker.js index d203c03..36a5899 100644 --- a/test2/server/services/contextKeywordTracker.js +++ b/test2/server/services/contextKeywordTracker.js @@ -85,23 +85,31 @@ class ContextKeywordTracker { return data.keywords; } - enrichQueryWithContext(sessionId, query) { + enrichQueryWithContext(sessionId, query, session = null) { const normalized = (query || '').trim(); - const keywords = this.getSessionKeywords(sessionId); - if (keywords.length === 0) { + const isSimpleFollowUp = /^(这个|那个|它|它的|他|他的|该|这款|那款|详细|继续|怎么|为什么|适合谁|什么意思|怎么吃|怎么用|功效|成分|多少钱|哪里买|价格|副作用|正规吗|地址|电话|联系方式|区别|哪个好|规格|包装|剂型|形态|一天几次|每天几次|每日几次)/i.test(normalized); + + if (!isSimpleFollowUp) { return normalized; } - const isSimpleFollowUp = /^(这个|那个|它|该|这款|那款|详细|继续|怎么|为什么|适合谁|什么意思|怎么吃|怎么用|功效|成分|多少钱|哪里买|价格|副作用|正规吗|地址|电话|联系方式|区别|哪个好)/i.test(normalized); - - if (isSimpleFollowUp) { - const keywordStr = keywords.slice(-3).join(' '); - console.log(`[ContextTracker] Enriching: "${normalized}" + "${keywordStr}"`); - return `${keywordStr} ${normalized}`; + // 优先用session的KB话题记忆（60秒内有效） + // 解决：聊了"一成系统"再聊"骨关节"后追问"这款怎么吃"，应关联"骨关节"而非"一成系统" + const KB_TOPIC_TTL = 60000; + if (session?._lastKbTopic && session?._lastKbHitAt && (Date.now() - session._lastKbHitAt < KB_TOPIC_TTL)) { + console.log(`[ContextTracker] Enriching from KB topic memory: "${normalized}" + "${session._lastKbTopic}"`); + return `${session._lastKbTopic} ${normalized}`; } - return normalized; + // fallback: 原有keyword tracker逻辑 + const keywords = this.getSessionKeywords(sessionId); + if (keywords.length === 0) { + return normalized; + } + const keywordStr = keywords.slice(-3).join(' '); + console.log(`[ContextTracker] Enriching: "${normalized}" + "${keywordStr}"`); + return `${keywordStr} ${normalized}`; } cleanup() { diff --git a/test2/server/services/fastAsrCorrector.js b/test2/server/services/fastAsrCorrector.js index e0c5c1a..5d8212c 100644 --- a/test2/server/services/fastAsrCorrector.js +++ b/test2/server/services/fastAsrCorrector.js @@ -31,6 +31,14 @@ const PHRASE_MAP = { '一陈系统': '一成系统', '依成系统': '一成系统', '伊成系统': '一成系统', + '益生系统': '一成系统', + '易诚系统': '一成系统', + '易乘系统': '一成系统', + '一声系统': '一成系统', + '亿生系统': '一成系统', + '义诚系统': '一成系统', + '忆诚系统': '一成系统', + '以诚系统': '一成系统', '盛咖学院': '盛咖学愿', '圣咖学愿': '盛咖学愿', '盛卡学愿': '盛咖学愿', @@ -65,6 +73,8 @@ const WORD_MAP = { '一乘': '一成', '一承': '一成', '一丞': '一成', '一呈': '一成', '一澄': '一成', '一橙': '一成', '一层': '一成', '一陈': '一成', '依成': '一成', '伊成': '一成', + '益生': '一成', '易诚': '一成', '义诚': '一成', '忆诚': '一成', '以诚': '一成', + '一声': '一成', '亿生': '一成', '易乘': '一成', '大窝': '大沃', '大握': '大沃', '大我': '大沃', '大卧': '大沃', '爱众享': 'Ai众享', '艾众享': 'Ai众享', '哎众享': 'Ai众享', '小洪': '小红', '小宏': '小红', '小鸿': '小红', @@ -123,7 +133,7 @@ function correctAsrText(text) { result = replaceOrderedMappings(result, WORD_MAP); // 激进策略：所有"X+系统"格式（非常见系统词）一律转为"一成系统" - result = result.replace(/[一二三四五六七八九十壹贰叁肆伍陆柒捌玖拾百千万亿兆零两几单双半多少全数整这那某每各以已亦艺毅怡逸溢义忆益伊依乙翼奕弈邑佚颐译蚁屹役疫裔翊熠旖漪倚绮峄羿轶壹弋驿奕懿肄翌苡圯佾诒铱仡]{1,2}(?:成|城|程|诚|乘|承|丞|呈|澄|橙|层|陈|趁|撑|称|秤|盛|剩|胜)系统/g, '一成系统'); + result = result.replace(/[一二三四五六七八九十壹贰叁肆伍陆柒捌玖拾百千万亿兆零两几单双半多少全数整这那某每各以已亦艺毅怡逸溢义忆益伊依乙翼奕弈邑佚颐译蚁屹役疫裔翊熠旖漪倚绮峄羿轶壹弋驿奕懿肄翌苡圯佾诒铱仡易]{1,2}(?:成|城|程|诚|乘|承|丞|呈|澄|橙|层|陈|趁|撑|称|秤|盛|剩|胜|生|声)系统/g, '一成系统'); for (const [from, to] of Object.entries(PRODUCT_ALIAS_MAP).sort((a, b) => b[0].length - a[0].length)) { if (shouldExpandProductAlias(result, from)) { diff --git a/test2/server/services/knowledgeKeywords.js b/test2/server/services/knowledgeKeywords.js index 031af2f..82a2140 100644 --- a/test2/server/services/knowledgeKeywords.js +++ b/test2/server/services/knowledgeKeywords.js @@ -222,6 +222,9 @@ const ROUTE_TOPIC_KEYWORDS = [ '慈善', '慈善事业', '社会责任', + 'Rolf Sorg', + 'RolfSorg', + '斯派尔', '不上市', '汽车奖励', '退休金', @@ -359,6 +362,8 @@ const ROUTE_TOPIC_KEYWORDS = [ '直销还是传销', '合不合法', '正不正规', + '正规吗', + '合法吗', '层级分销', '非法集资', '拉人头', @@ -410,6 +415,8 @@ const ROUTE_TOPIC_KEYWORDS = [ '区别', '哪个好', '多久见效', + '见效', + '多久能见效', '哪里买', '怎么买', '保质期', @@ -466,6 +473,110 @@ const ROUTE_TOPIC_KEYWORDS = [ '搭配吃', '吃药', '药物', + // 产品剂型/形态（用户质疑/纠正时常提及） + '粉末', + '粉剂', + '粉状', + '冲剂', + '冲泡', + '片剂', + '药片', + '胶囊', + '软胶囊', + '颗粒', + '口服液', + '膏状', + // 质疑/纠正/确认/怀疑/复查类口语词（全覆盖） + // 直接否定 + '不是的', + '才不是', + '不是不是', + '不是这么回事', + // 指出错误 + '搞错了', + '说错了', + '弄错了', + '记错了', + '搞混了', + '搞反了', + '记岔了', + '说反了', + '张冠李戴', + '答非所问', + // 说AI不对 + '不对', + '不是这样', + '不准确', + '不正确', + '有误', + '说的不对', + '回答有误', + '不太对', + '不太准', + // 与认知矛盾 + '不一样', + '不一致', + '前后矛盾', + '自相矛盾', + // 怀疑/不信 + '不信', + '骗人', + '忽悠', + '吹牛', + '太夸张', + '离谱', + '扯淡', + '瞎扯', + // 确认/复查 + '你确定吗', + '确定吗', + '真的吗', + '当真', + '再查一下', + '再确认一下', + '再核实', + '重新查', + '核实一下', + '查清楚', + '搞清楚', + // 委婉质疑 + '好像不是', + '好像不对', + '我觉得不对', + '恐怕不是', + '感觉不对', + // 质问来源 + '谁说的', + '谁告诉你', + '有什么根据', + '有什么依据', + '有证据吗', + '有依据吗', + // 不可能/反问 + '怎么可能', + '不可能', + '不会吧', + '不是吧', + '开玩笑', + '别逗了', + '胡说', + '瞎说', + '乱说', + // 纠正句式 + '到底是', + '究竟是', + '应该是', + '明明是', + '其实是', + '本来是', + '怎么变成', + '不应该是', + // 产品形态/使用方式 + '冲着喝', + '泡着喝', + '直接吞', + '是喝的', + '是吃的', ]; const CANONICAL_KNOWLEDGE_TERMS = [ @@ -896,6 +1007,7 @@ const SCIENCE_TRAINING_ROUTE_KEYWORDS = uniqueKeywords([ const KNOWLEDGE_ROUTE_KEYWORDS = uniqueKeywords([ ...KNOWLEDGE_ENTITY_KEYWORDS, ...ROUTE_TOPIC_KEYWORDS, + ...FAQ_ROUTE_KEYWORDS, ]); const TRACKER_KEYWORD_GROUPS = [ diff --git a/test2/server/services/nativeVoiceGateway.js b/test2/server/services/nativeVoiceGateway.js index 01fe624..0cc62e3 100644 --- a/test2/server/services/nativeVoiceGateway.js +++ b/test2/server/services/nativeVoiceGateway.js @@ -29,6 +29,7 @@ const { buildVoiceSystemRole, buildVoiceGreeting, } = require('./assistantProfileConfig'); +const { getAssistantProfile } = require('./assistantProfileService'); const sessions = new Map(); @@ -498,6 +499,10 @@ async function processReply(session, text, turnSeq = session.latestUserTurnSeq | session._lastKbTopic = cleanText; session._lastKbHitAt = Date.now(); } + // 直接用KB原始回答作为字幕，不依赖S2S event 351（S2S可能拆段/改写/丢失内容） + const ragSubtitleText = ragContent.map((item) => item.content).join(' '); + persistAssistantSpeech(session, ragSubtitleText, { source, toolName, meta: responseMeta }); + session.lastDeliveredAssistantTurnSeq = activeTurnSeq; session._pendingExternalRagReply = true; await sendExternalRag(session, ragContent); session.awaitingUpstreamReply = true; @@ -1003,7 +1008,10 @@ function attachClientHandlers(session) { } if (parsed.type === 'start') { + session.userId = parsed.userId || session.userId || null; + const remoteProfileResult = await getAssistantProfile({ userId: session.userId }); const assistantProfile = resolveAssistantProfile({ + ...(remoteProfileResult.profile || {}), ...(session.assistantProfile || {}), ...((parsed.assistantProfile && typeof parsed.assistantProfile === 'object') ? parsed.assistantProfile : {}), }); @@ -1014,7 +1022,6 @@ function attachClientHandlers(session) { session.speaker = parsed.speaker || process.env.VOLC_S2S_SPEAKER_ID || 'zh_female_vv_jupiter_bigtts'; session.modelVersion = parsed.modelVersion || 'O'; session.greetingText = parsed.greetingText || buildVoiceGreeting(assistantProfile); - session.userId = parsed.userId || session.userId || null; // 立即发送 ready，不等 upstream event 150，大幅缩短前端等待时间 sendReady(session); session.upstream = createUpstreamConnection(session); diff --git a/test2/server/services/toolExecutor.js b/test2/server/services/toolExecutor.js index aee15c0..80d47ed 100644 --- a/test2/server/services/toolExecutor.js +++ b/test2/server/services/toolExecutor.js @@ -1,7 +1,8 @@ const axios = require('axios'); const https = require('https'); const arkChatService = require('./arkChatService'); -const { buildKnowledgeAnswerPrompt } = require('./assistantProfileConfig'); +const { buildKnowledgeAnswerPrompt, resolveAssistantProfile } = require('./assistantProfileConfig'); +const { getAssistantProfile } = require('./assistantProfileService'); // HTTP keep-alive agent：复用TCP连接，避免每次请求重新握手 const kbHttpAgent = new https.Agent({ @@ -43,14 +44,14 @@ const { SCIENCE_TRAINING_ROUTE_KEYWORDS, } = require('./knowledgeKeywords'); -// KB查询缓存：相同effectiveQuery + datasetIds在TTL内直接返回缓存结果 +// KB查询缓存：相同effectiveQuery + datasetIds + userId在TTL内直接返回缓存结果 const KB_CACHE_TTL_MS = 5 * 60 * 1000; // 5分钟 (hit结果) const KB_CACHE_NOHIT_TTL_MS = 2 * 60 * 1000; // 2分钟 (no-hit结果，较短TTL) const KB_CACHE_MAX_SIZE = 200; const kbQueryCache = new Map(); -function getKbCacheKey(query, datasetIds) { - return `${(query || '').trim()}|${(datasetIds || []).sort().join(',')}`; +function getKbCacheKey(query, datasetIds, profileScope = 'global') { + return `${String(profileScope || 'global').trim() || 'global'}|${(query || '').trim()}|${(datasetIds || []).sort().join(',')}`; } function getKbCache(key) { @@ -698,6 +699,16 @@ class ToolExecutor { query = query || ''; const responseMode = response_mode === 'snippet' ? 'snippet' : 'answer'; const knowledgeEndpointId = process.env.VOLC_ARK_KNOWLEDGE_ENDPOINT_ID || process.env.VOLC_ARK_ENDPOINT_ID; + const profileUserId = _session?.profileUserId || _session?.userId || null; + const assistantProfileResult = await getAssistantProfile({ userId: profileUserId }); + const assistantProfile = resolveAssistantProfile({ + ...(assistantProfileResult?.profile || {}), + ...(_session?.assistantProfile || {}), + }); + if (_session && assistantProfileResult?.profile) { + _session.assistantProfile = assistantProfile; + } + const profileScope = profileUserId || 'global'; console.log(`[ToolExecutor] searchKnowledge called with query="${query}"`); // 注意：answer 模式必须依据知识库回答，因此不再允许本地热答案直接绕过知识库。 @@ -748,7 +759,7 @@ class ToolExecutor { } try { // 缓存检查：相同effectiveQuery + datasetIds命中缓存时直接返回，避免重复API调用 - const cacheKey = getKbCacheKey(effectiveQuery, kbTarget.datasetIds); + const cacheKey = getKbCacheKey(effectiveQuery, kbTarget.datasetIds, profileScope); const cached = getKbCache(cacheKey); if (cached) { const latencyMs = Date.now() - startTime; @@ -764,7 +775,7 @@ class ToolExecutor { }; } console.log('[ToolExecutor] Trying Ark Knowledge Search...'); - const arkResult = await this.searchArkKnowledge(effectiveQuery, context, responseMode, kbTarget.datasetIds, query, _session?.assistantProfile || null); + const arkResult = await this.searchArkKnowledge(effectiveQuery, context, responseMode, kbTarget.datasetIds, query, assistantProfile); const latencyMs = Date.now() - startTime; console.log(`[ToolExecutor] Ark KB search succeeded in ${latencyMs}ms`); // 缓存所有结果（hit用5分钟TTL，no-hit用2分钟TTL），避免重复API调用 diff --git a/test2/server/tests/README_VIKING_TEST.md b/test2/server/tests/README_VIKING_TEST.md new file mode 100644 index 0000000..49e5cb0 --- /dev/null +++ b/test2/server/tests/README_VIKING_TEST.md @@ -0,0 +1,122 @@ +# Viking 检索性能测试套件 + +本测试套件用于测试火山引擎（Viking）知识库的检索性能，包括延迟、缓存效率、并发吞吐量等指标。 + +## 目录结构 + +``` +test2/server/tests/ +├── viking_retrieval_performance.js # 核心性能测试类 +├── quick_test_viking.js # 快速测试脚本 +├── test_results/ # 测试结果输出目录 +└── README_VIKING_TEST.md # 本文档 +``` + +## 使用方法 + +### 前置条件 + +1. 确保已安装依赖： +```bash +cd test2/server +npm install +``` + +2. 确保 `.env` 文件配置正确，包含火山引擎相关环境变量： +``` +VOLC_ARK_API_KEY=your_api_key +VOLC_ARK_ENDPOINT_ID=your_endpoint_id +VOLC_ARK_KNOWLEDGE_BASE_IDS=your_kb_ids +VOLC_ARK_KNOWLEDGE_ENDPOINT_ID=your_kb_endpoint_id +``` + +### 快速测试 + +运行完整测试套件： +```bash +cd test2/server +node tests/quick_test_viking.js +``` + +运行特定类型的测试： + +```bash +# 只运行延迟测试 +node tests/quick_test_viking.js latency + +# 只运行缓存效率测试 +node tests/quick_test_viking.js cache + +# 只运行并发测试 +node tests/quick_test_viking.js concurrency +``` + +### 编程使用 + +在你的代码中集成性能测试： + +```javascript +const VikingRetrievalPerformanceTester = require('./tests/viking_retrieval_performance'); + +const tester = new VikingRetrievalPerformanceTester({ + outputDir: './my_test_results', + verbose: true, + warmupRuns: 3 +}); + +// 运行完整测试套件 +await tester.runFullSuite(); + +// 或者单独运行测试 +const latencyResults = await tester.testLatency([ + { name: 'My Query', query: '测试查询' } +], 10); + +// 生成并保存报告 +tester.printSummary(); +tester.saveReport('my_test.json'); +``` + +## 测试类型 + +### 1. 延迟测试 (Latency Test) +测试不同查询的响应时间，包括： +- 平均延迟 +- P50/P95/P99 延迟 +- 最小/最大延迟 +- 命中率 + +### 2. 缓存效率测试 (Cache Efficiency Test) +测试缓存命中时的性能提升： +- 首次查询延迟 +- 缓存命中延迟 +- 加速比（Speedup） + +### 3. 并发测试 (Concurrency Test) +测试不同并发级别下的吞吐量： +- 吞吐量（requests/second） +- 成功率 +- 总耗时 + +### 4. 查询类型测试 (Query Types Test) +测试不同类型查询的性能差异。 + +## 输出结果 + +测试结果将保存为 JSON 文件，包含： +- 完整的测试数据 +- 摘要统计 +- 时间戳 + +控制台会输出格式化的测试摘要。 + +## 示例测试查询 + +默认测试查询包括： +- 产品查询（小红、大白） +- 公司信息 +- NTC 技术 +- 高频问题 +- 无结果查询 + +你可以根据自己的知识库内容自定义测试查询。 diff --git a/test2/server/tests/VIKING_PERFORMANCE_REPORT.md b/test2/server/tests/VIKING_PERFORMANCE_REPORT.md new file mode 100644 index 0000000..7ca1930 --- /dev/null +++ b/test2/server/tests/VIKING_PERFORMANCE_REPORT.md @@ -0,0 +1,111 @@ +# Viking 检索性能测试报告 + +## 测试日期 +2026-03-20 + +## 测试环境 +- 项目: bigwo/test2/server +- 测试文件: test_viking_direct_api.js +- 测试方法: 直接调用火山引擎方舟API + +## 测试结果 + +### 直接API测试（无查询改写，无缓存） + +| 查询名称 | 平均延迟 | P50延迟 | P95延迟 | P99延迟 | 最小延迟 | 最大延迟 | +|---------|---------|---------|---------|---------|---------|---------| +| CC胶囊 Direct | 3098.06ms | 4639.62ms | 8949.48ms | 8949.48ms | 1744.93ms | 4639.62ms | +| IB5 Direct | 4130.82ms | 4639.62ms | 8949.48ms | 8949.48ms | 2567.20ms | 6941.14ms | +| 邓白氏 Direct | 4607.89ms | 4639.62ms | 8949.48ms | 8949.48ms | 3486.05ms | 6355.73ms | +| Q10 Direct | 5156.85ms | 4639.62ms | 8949.48ms | 8949.48ms | 4146.50ms | 6264.39ms | +| 火炉原理 Direct | 7557.88ms | 4639.62ms | 8949.48ms | 8949.48ms | 5917.74ms | 8949.48ms | + +### 总体统计 + +| 指标 | 数值 | +|------|------| +| 总体平均延迟 | 4910.30ms | +| 总体P50延迟 | 4639.62ms | +| 总体P95延迟 | 8949.48ms | +| 总体P99延迟 | 8949.48ms | +| 总体最小延迟 | 1744.93ms | +| 总体最大延迟 | 8949.48ms | + +### 冷启动测试（首次调用） + +| 查询名称 | 首次延迟 | +|---------|---------| +| Q10 Unique | 5770.73ms | +| IB5 Unique | 5389.67ms | +| CC胶囊 Unique | 5079.27ms | +| 邓白氏 Unique | 5069.32ms | +| 火炉原理 Unique | 5669.52ms | + +**首次调用平均延迟**: 5395.70ms + +### 缓存命中测试 + +| 场景 | 延迟 | 加速比 | +|------|------|--------| +| 高频问题 (HOT_ANSWER) | ~0.15ms | ~35000x | +| 知识库缓存 (Ark KB Cache) | ~1-2ms | ~2500x | + +## 性能分析 + +### 1. 原始API调用延迟 +- **平均**: ~4.9秒 +- **P50**: ~4.6秒 +- **P95**: ~8.9秒 + +### 2. 缓存优化效果 +项目中的多层缓存机制带来了显著的性能提升： + +1. **高频问题缓存**: ~0.15ms，提升约35,000倍 +2. **知识库结果缓存**: ~1-2ms，提升约2,500倍 +3. **查询改写 + 缓存**: 进一步提升命中率 + +### 3. 各层延迟分布 + +``` +真实API调用: ~4.9秒 + ↓ +知识库缓存: ~1-2ms (提升2500x) + ↓ +高频问题缓存: ~0.15ms (提升35000x) +``` + +## 测试文件 + +本次测试使用的文件： +1. `viking_retrieval_performance.js` - 完整测试套件 +2. `viking_retrieval_performance_with_mock.js` - 带模拟模式的测试 +3. `test_real_viking_kb.js` - 真实知识库测试 +4. `test_viking_cold_start.js` - 冷启动测试 +5. `test_viking_direct_api.js` - 直接API测试 +6. `quick_test_viking.js` - 快速测试脚本 +7. `run_real_test.js` - 自动检测配置测试 + +## 结论 + +1. **原始Viking API延迟**: 约4-9秒 +2. **缓存优化效果显著**: 多层缓存可将延迟降低到毫秒级 +3. **查询改写机制**: 有效提升缓存命中率 +4. **推荐配置**: + - 保持当前的缓存策略 + - 考虑增加高频问题的覆盖范围 + - 监控P95延迟，优化长尾请求 + +## 使用方法 + +```bash +cd test2/server + +# 运行完整测试（模拟模式） +node tests/viking_retrieval_performance_with_mock.js + +# 运行真实测试 +node tests/test_viking_direct_api.js + +# 快速测试 +node tests/quick_test_viking.js +``` diff --git a/test2/server/tests/_deploy_kb_prompt.cjs b/test2/server/tests/_deploy_kb_prompt.cjs new file mode 100644 index 0000000..734796e --- /dev/null +++ b/test2/server/tests/_deploy_kb_prompt.cjs @@ -0,0 +1,95 @@ +const { Client } = require('ssh2'); +const fs = require('fs'); +const path = require('path'); + +const SERVER = { host: '119.45.10.34', port: 22, username: 'root', password: '#xyzh%CS#2512@28' }; +const REMOTE_DIR = '/www/wwwroot/demo.tensorgrove.com.cn/server'; + +const FILES = [ + { + local: path.join(__dirname, '..', 'services', 'assistantProfileConfig.js'), + remote: REMOTE_DIR + '/services/assistantProfileConfig.js', + }, +]; + +function sshExec(client, cmd) { + return new Promise((resolve, reject) => { + client.exec(cmd, (err, stream) => { + if (err) return reject(err); + let out = '', errOut = ''; + stream.on('data', (d) => (out += d.toString())); + stream.stderr.on('data', (d) => (errOut += d.toString())); + stream.on('close', (code) => resolve({ out: out.trim(), err: errOut.trim(), code })); + }); + }); +} + +function sshUpload(client, localPath, remotePath) { + return new Promise((resolve, reject) => { + client.sftp((err, sftp) => { + if (err) return reject(err); + sftp.fastPut(localPath, remotePath, (err) => { + if (err) return reject(err); + resolve(); + }); + }); + }); +} + +async function deploy() { + const client = new Client(); + client.on('ready', async () => { + try { + console.log('✅ SSH connected\n'); + const ts = new Date().toISOString().replace(/[:.]/g, '-').slice(0, 19); + + // 1. 先查看服务器当前prompt + console.log('📋 服务器当前prompt:'); + const before = await sshExec(client, `grep -n 'buildKnowledgeAnswerPrompt\\|知识库涵盖\\|产品用法' ${REMOTE_DIR}/services/assistantProfileConfig.js | head -5`); + console.log(' ' + (before.out || '(not found)') + '\n'); + + // 2. 备份 + 上传 + for (const { local, remote } of FILES) { + const name = path.basename(remote); + await sshExec(client, `cp ${remote} ${remote}.bak_${ts}`); + console.log('📦 Backup: ' + name); + await sshUpload(client, local, remote); + console.log('📤 Uploaded: ' + name); + const syntax = await sshExec(client, `node -c ${remote}`); + if (syntax.code !== 0) { + console.error('❌ Syntax error in ' + name + '! Rolling back...'); + await sshExec(client, `cp ${remote}.bak_${ts} ${remote}`); + client.end(); + return; + } + console.log('🔍 Syntax OK: ' + name + '\n'); + } + + // 3. 重启PM2 + const pm2Result = await sshExec(client, `cd ${REMOTE_DIR} && pm2 restart all --update-env`); + console.log('🔄 PM2 restarted'); + if (pm2Result.out) console.log(' ' + pm2Result.out.split('\n').slice(0, 3).join('\n ')); + await new Promise(r => setTimeout(r, 5000)); + + // 4. 验证新prompt已生效 + console.log('\n📋 服务器新prompt:'); + const after = await sshExec(client, `grep -n '知识库涵盖\\|产品常有别名\\|不得编造' ${REMOTE_DIR}/services/assistantProfileConfig.js | head -5`); + console.log(' ' + (after.out || '(not found)')); + + // 5. PM2 状态 + const status = await sshExec(client, 'pm2 status'); + console.log('\n📊 PM2 Status:'); + console.log(' ' + status.out.split('\n').slice(0, 6).join('\n ')); + + console.log('\n✅ 部署完成！KB answer prompt 已更新为优化版本'); + } catch (e) { + console.error('❌ Error:', e.message); + } finally { + client.end(); + } + }); + client.on('error', (err) => console.error('❌ SSH Error:', err.message)); + client.connect(SERVER); +} + +deploy(); diff --git a/test2/server/tests/_test_single.js b/test2/server/tests/_test_single.js new file mode 100644 index 0000000..fd08fd6 --- /dev/null +++ b/test2/server/tests/_test_single.js @@ -0,0 +1,70 @@ +const path = require('path'); +const https = require('https'); +const fs = require('fs'); + +// 加载 .env +const envPath = path.join(__dirname, '../.env'); +if (fs.existsSync(envPath)) { + fs.readFileSync(envPath, 'utf8').split('\n').forEach(line => { + const t = line.trim(); + if (!t || t.startsWith('#')) return; + const i = t.indexOf('='); + if (i > 0) { + const k = t.slice(0, i).trim(); + let v = t.slice(i + 1).trim(); + if ((v.startsWith('"') && v.endsWith('"')) || (v.startsWith("'") && v.endsWith("'"))) + v = v.slice(1, -1); + if (!process.env[k]) process.env[k] = v; + } + }); +} + +const PROMPT = '你是大沃的智能助手，负责回答与PM-FitLine德国产品相关的问题。知识库涵盖近50款PM-FitLine产品的完整资料（成分、用法、剂量、价格、规格、搭配方案、好转反应等）及117个常见问答。回答规则：产品相关具体信息必须严格依据知识库，不得猜测或自行补充；公司背景、健康常识可适当补充。产品常有别名（小红=艾特维、大白=倍适、小白=维适多等），请注意识别。不得编造产品名或数据。PM是营养品非药物，涉及疾病建议咨询医生。若知识库无相关内容，坦诚说明并建议咨询推荐人。回答口语化、简洁，1-3句给结论，150字内。'; + +const QUERY = process.argv[2] || '一成系统 赋能 德国PM'; + +const m = process.env.VOLC_ARK_KB_MODEL || process.env.VOLC_ARK_KNOWLEDGE_ENDPOINT_ID; +const ak = process.env.VOLC_ARK_API_KEY || process.env.VOLC_ACCESS_KEY_ID; +const ids = (process.env.VOLC_ARK_KNOWLEDGE_BASE_IDS || '').split(',').map(s => s.trim()).filter(Boolean); +const th = parseFloat(process.env.VOLC_ARK_KNOWLEDGE_THRESHOLD) || 0.3; + +const body = JSON.stringify({ + model: m, + messages: [ + { role: 'system', content: PROMPT }, + { role: 'user', content: QUERY }, + ], + metadata: { knowledge_base: { dataset_ids: ids, top_k: 3, threshold: th } }, + stream: false, + max_tokens: 80, + thinking: { type: 'disabled' }, +}); + +const t0 = Date.now(); +const req = https.request({ + hostname: 'ark.cn-beijing.volces.com', + path: '/api/v3/chat/completions', + method: 'POST', + headers: { + 'Content-Type': 'application/json', + 'Authorization': `Bearer ${ak}`, + 'Content-Length': Buffer.byteLength(body), + }, + timeout: 15000, +}, (res) => { + let chunks = []; + res.on('data', c => chunks.push(c)); + res.on('end', () => { + const j = JSON.parse(Buffer.concat(chunks).toString()); + const content = j.choices?.[0]?.message?.content || '(empty)'; + const u = j.usage || {}; + const ms = Date.now() - t0; + const isNoHit = /未找到|未提及|暂无|没有相关|无法|不确定/.test(content); + console.log(`\n检索词: "${QUERY}"`); + console.log(`延迟: ${ms}ms | tokens: ${u.prompt_tokens}->${u.completion_tokens} | ${isNoHit ? '❌NO-HIT' : '✅HIT'}`); + console.log(`\n回答:\n${content}\n`); + }); +}); +req.on('error', e => console.error('ERROR:', e.message)); +req.write(body); +req.end(); diff --git a/test2/server/tests/quick_test_viking.js b/test2/server/tests/quick_test_viking.js new file mode 100644 index 0000000..c8f5647 --- /dev/null +++ b/test2/server/tests/quick_test_viking.js @@ -0,0 +1,59 @@ +const path = require('path'); +const VikingRetrievalPerformanceTester = require('./viking_retrieval_performance'); + +require('dotenv').config({ path: path.join(__dirname, '../.env') }); + +async function main() { + const args = process.argv.slice(2); + const testType = args[0] || 'full'; + + const tester = new VikingRetrievalPerformanceTester(); + + const testQueries = [ + { name: 'Product Query - Xiaohong', query: '小红产品有什么功效' }, + { name: 'Product Query - Dabai', query: '大白产品怎么吃' }, + { name: 'Company Info', query: '德国PM公司介绍' }, + { name: 'NTC Technology', query: 'NTC营养保送系统原理' }, + { name: 'Hot Answer', query: '基础三合一怎么吃' }, + { name: 'No Hit Query', query: '今天天气怎么样' } + ]; + + switch (testType) { + case 'latency': + console.log('Running latency test only...'); + await tester.warmup(testQueries.map(q => q.query)); + await tester.testLatency(testQueries, 5); + tester.printSummary(); + tester.saveReport('latency_test.json'); + break; + + case 'cache': + console.log('Running cache efficiency test only...'); + await tester.warmup(testQueries.map(q => q.query)); + await tester.testCacheEfficiency(testQueries.slice(0, 3), 5); + tester.printSummary(); + tester.saveReport('cache_test.json'); + break; + + case 'concurrency': + console.log('Running concurrency test only...'); + await tester.warmup(testQueries.map(q => q.query)); + await tester.testConcurrency(testQueries.slice(0, 3), [1, 2, 3, 5]); + tester.printSummary(); + tester.saveReport('concurrency_test.json'); + break; + + case 'full': + default: + console.log('Running full performance test suite...'); + await tester.runFullSuite(); + break; + } + + console.log('\nTest completed!'); +} + +main().catch(err => { + console.error('Test failed:', err); + process.exit(1); +}); diff --git a/test2/server/tests/run_real_test.js b/test2/server/tests/run_real_test.js new file mode 100644 index 0000000..e132c6b --- /dev/null +++ b/test2/server/tests/run_real_test.js @@ -0,0 +1,68 @@ +const path = require('path'); + +require('dotenv').config({ path: path.join(__dirname, '../.env') }); + +console.log('Checking environment variables...'); +console.log('VOLC_ARK_API_KEY:', process.env.VOLC_ARK_API_KEY ? 'Set' : 'Not set'); +console.log('VOLC_ARK_ENDPOINT_ID:', process.env.VOLC_ARK_ENDPOINT_ID); +console.log('VOLC_ARK_KNOWLEDGE_BASE_IDS:', process.env.VOLC_ARK_KNOWLEDGE_BASE_IDS); +console.log('VOLC_ARK_KNOWLEDGE_ENDPOINT_ID:', process.env.VOLC_ARK_KNOWLEDGE_ENDPOINT_ID); + +const hasRequiredConfig = process.env.VOLC_ARK_API_KEY && + process.env.VOLC_ARK_ENDPOINT_ID && + process.env.VOLC_ARK_KNOWLEDGE_BASE_IDS && + process.env.VOLC_ARK_KNOWLEDGE_ENDPOINT_ID; + +if (hasRequiredConfig) { + console.log('\nAll required configs found! Running real test...\n'); + + const VikingTester = require('./viking_retrieval_performance_with_mock'); + + (async () => { + try { + const tester = new VikingTester({ mockMode: false }); + await tester.runFullSuite(); + } catch (err) { + console.error('Test failed:', err); + } + })(); +} else { + console.log('\nMissing required environment variables. Checking parent directory...'); + + require('dotenv').config({ path: path.join(__dirname, '../../.env') }); + + console.log('VOLC_ARK_API_KEY (parent):', process.env.VOLC_ARK_API_KEY ? 'Set' : 'Not set'); + + const hasRequiredConfigParent = process.env.VOLC_ARK_API_KEY && + process.env.VOLC_ARK_ENDPOINT_ID && + process.env.VOLC_ARK_KNOWLEDGE_BASE_IDS && + process.env.VOLC_ARK_KNOWLEDGE_ENDPOINT_ID; + + if (hasRequiredConfigParent) { + console.log('\nConfigs found in parent directory! Running real test...\n'); + + const VikingTester = require('./viking_retrieval_performance_with_mock'); + + (async () => { + try { + const tester = new VikingTester({ mockMode: false }); + await tester.runFullSuite(); + } catch (err) { + console.error('Test failed:', err); + } + })(); + } else { + console.log('\nNo configs found. Running mock test instead...\n'); + + const VikingTester = require('./viking_retrieval_performance_with_mock'); + + (async () => { + try { + const tester = new VikingTester({ mockMode: true }); + await tester.runFullSuite(); + } catch (err) { + console.error('Test failed:', err); + } + })(); + } +} diff --git a/test2/server/tests/test_asr_coverage.js b/test2/server/tests/test_asr_coverage.js new file mode 100644 index 0000000..8030360 --- /dev/null +++ b/test2/server/tests/test_asr_coverage.js @@ -0,0 +1,520 @@ +/** + * 语音识别(ASR)纠错全覆盖测试 + * 覆盖：PHRASE_MAP、WORD_MAP、PRODUCT_ALIAS_MAP、激进正则、normalizeKnowledgeAlias、组合流水线 + * + * 运行方式: node --test tests/test_asr_coverage.js + */ +const { describe, it } = require('node:test'); +const assert = require('node:assert/strict'); + +const { correctAsrText, PHRASE_MAP, WORD_MAP, PRODUCT_ALIAS_MAP } = require('../services/fastAsrCorrector'); +const { normalizeKnowledgeAlias, shouldForceKnowledgeRoute } = require('../services/realtimeDialogRouting'); +const { hasKnowledgeRouteKeyword } = require('../services/knowledgeKeywords'); + +// 辅助：完整ASR流水线（模拟nativeVoiceGateway.extractUserText + routing） +function fullAsrPipeline(rawText) { + const corrected = correctAsrText(rawText); + const normalized = normalizeKnowledgeAlias(corrected); + return normalized; +} + +function assertPipelineRouteKb(rawText, msg) { + const processed = fullAsrPipeline(rawText); + const result = shouldForceKnowledgeRoute(processed); + assert.equal(result, true, msg || `ASR "${rawText}" → processed "${processed}" should route to KB`); +} + +// ================================================================ +// 1. PHRASE_MAP 全覆盖 —— 每条短语映射逐一验证 +// ================================================================ +describe('PHRASE_MAP —— 短语级ASR纠错全覆盖', () => { + + describe('一成系统变体（36条）', () => { + const yichengVariants = [ + '一城系统', '逸城系统', '一程系统', '易成系统', '一诚系统', + '亦成系统', '艺成系统', '溢成系统', '义成系统', '毅成系统', + '怡成系统', '以成系统', '已成系统', '亿成系统', '忆成系统', + '益成系统', '一乘系统', '一承系统', '一丞系统', '一呈系统', + '一澄系统', '一橙系统', '一层系统', '一趁系统', '一陈系统', + '依成系统', '伊成系统', '益生系统', '易诚系统', '易乘系统', + '一声系统', '亿生系统', '义诚系统', '忆诚系统', '以诚系统', + ]; + + for (const variant of yichengVariants) { + it(`"${variant}" → "一成系统"`, () => { + const result = correctAsrText(variant); + assert.ok(result.includes('一成系统'), `"${variant}" should correct to 一成系统, got "${result}"`); + }); + } + }); + + describe('其他短语映射', () => { + const otherPhrases = [ + ['盛咖学院', '盛咖学愿'], + ['圣咖学愿', '盛咖学愿'], + ['盛卡学愿', '盛咖学愿'], + ['营养配送系统', 'NTC营养保送系统'], + ['营养输送系统', 'NTC营养保送系统'], + ['营养传送系统', 'NTC营养保送系统'], + ['营养传输系统', 'NTC营养保送系统'], + ['暖炉原理', '火炉原理'], + ['整应反应', '好转反应'], + ['整健反应', '好转反应'], + ['排毒反应', '好转反应'], + ['5加1', '5+1'], + ['五加一', '5+1'], + ['起步三观', '起步三关'], + ['起步三官', '起步三关'], + ['doublepm', '德国PM'], + ['double pm', '德国PM'], + ['DoublePM', '德国PM'], + ['Double PM', '德国PM'], + ['DOUBLEPM', '德国PM'], + ['DOUBLE PM', '德国PM'], + ['基础三合一', 'PM细胞营养素 基础套装'], + ['三合一基础套', 'PM细胞营养素 基础套装'], + ['大白小红小白', 'PM细胞营养素 基础套装'], + ]; + + for (const [input, expected] of otherPhrases) { + it(`"${input}" → 含"${expected}"`, () => { + const result = correctAsrText(input); + assert.ok(result.includes(expected), `"${input}" should contain "${expected}", got "${result}"`); + }); + } + }); +}); + +// ================================================================ +// 2. WORD_MAP 全覆盖 —— 单词级ASR纠错 +// ================================================================ +describe('WORD_MAP —— 单词级ASR纠错全覆盖', () => { + + describe('一成/一城等同音变体', () => { + const wordVariants = [ + ['一城', '一成'], ['逸城', '一成'], ['一程', '一成'], ['易成', '一成'], + ['一诚', '一成'], ['亦成', '一成'], ['艺成', '一成'], ['溢成', '一成'], + ['义成', '一成'], ['毅成', '一成'], ['怡成', '一成'], ['以成', '一成'], + ['已成', '一成'], ['亿成', '一成'], ['忆成', '一成'], ['益成', '一成'], + ['一乘', '一成'], ['一承', '一成'], ['一丞', '一成'], ['一呈', '一成'], + ['一澄', '一成'], ['一橙', '一成'], ['一层', '一成'], ['一陈', '一成'], + ['依成', '一成'], ['伊成', '一成'], + ['益生', '一成'], ['易诚', '一成'], ['义诚', '一成'], ['忆诚', '一成'], ['以诚', '一成'], + ['一声', '一成'], ['亿生', '一成'], ['易乘', '一成'], + ]; + + for (const [input, expected] of wordVariants) { + it(`"${input}" → "${expected}"`, () => { + const result = correctAsrText(`${input}的介绍`); + assert.ok(result.includes(expected), `"${input}" should correct to "${expected}", got "${result}"`); + }); + } + }); + + describe('大沃同音变体', () => { + const dawoVariants = ['大窝', '大握', '大我', '大卧']; + for (const v of dawoVariants) { + it(`"${v}" → "大沃"`, () => { + const result = correctAsrText(v); + assert.ok(result.includes('大沃'), `"${v}" should correct to 大沃, got "${result}"`); + }); + } + }); + + describe('Ai众享同音变体', () => { + const aiVariants = ['爱众享', '艾众享', '哎众享']; + for (const v of aiVariants) { + it(`"${v}" → "Ai众享"`, () => { + const result = correctAsrText(v); + assert.ok(result.includes('Ai众享'), `"${v}" should correct to Ai众享, got "${result}"`); + }); + } + }); + + describe('产品名同音变体', () => { + const productVariants = [ + ['小洪', '小红'], ['小宏', '小红'], ['小鸿', '小红'], + ['大百', '大白'], ['大柏', '大白'], + ['小百', '小白'], ['小柏', '小白'], ['维适多', '小白'], + ]; + for (const [input, expected] of productVariants) { + it(`"${input}" → "${expected}"`, () => { + const result = correctAsrText(`${input}产品功效`); + assert.ok(result.includes(expected), `"${input}" should correct to "${expected}", got "${result}"`); + }); + } + }); + + describe('其他同音变体', () => { + const others = [ + ['营养配送', '营养保送'], + ['营养输送', '营养保送'], + ['阿玉吠陀', '阿育吠陀'], + ['阿育费陀', '阿育吠陀'], + ]; + for (const [input, expected] of others) { + it(`"${input}" → "${expected}"`, () => { + const result = correctAsrText(input); + assert.ok(result.includes(expected), `"${input}" should correct to "${expected}", got "${result}"`); + }); + } + }); +}); + +// ================================================================ +// 3. PRODUCT_ALIAS_MAP —— 产品别名扩展 +// ================================================================ +describe('PRODUCT_ALIAS_MAP —— 产品别名扩展', () => { + const aliasCases = [ + ['小红怎么吃', 'Activize'], + ['小红功效', 'Activize'], + ['Activize是什么', 'Activize Oxyplus'], + ['大白怎么吃', 'Basics'], + ['大白功效', 'Basics'], + ['Basics成分', 'Basics'], + ['小白怎么吃', 'Restorate'], + ['Restorate功效', 'Restorate'], + ['FitLine是什么', 'PM-FitLine'], + ['PM FitLine的功效', 'PM-FitLine'], + ['PM细胞营养', 'PM细胞营养素'], + ['PM营养素功效', 'PM细胞营养素'], + ['德国PM营养素', 'PM细胞营养素'], + ]; + + for (const [input, expectContain] of aliasCases) { + it(`"${input}" → 扩展含"${expectContain}"`, () => { + const result = correctAsrText(input); + assert.ok(result.includes(expectContain), `"${input}" should expand to contain "${expectContain}", got "${result}"`); + }); + } + + it('非追问位置不应触发产品扩展（如"小红帽"）', () => { + const result = correctAsrText('小红帽故事'); + // 小红帽 doesn't match the expansion pattern because 帽 is not in the suffix list + assert.ok(!result.includes('Activize'), `"小红帽故事" should NOT expand 小红, got "${result}"`); + }); +}); + +// ================================================================ +// 4. 激进正则 —— X+成/城/程...+系统 统一纠正 +// ================================================================ +describe('激进正则 —— 未在字典中的"X成系统"变体', () => { + const aggressiveCases = [ + '翼成系统', '奕成系统', '弈成系统', '颐成系统', + '译成系统', '蚁成系统', '壹成系统', + '一盛系统', '一胜系统', '一生系统', + '一称系统', '一撑系统', + '双成系统', '半成系统', + ]; + + for (const variant of aggressiveCases) { + it(`"${variant}" → "一成系统"（激进正则兜底）`, () => { + const result = correctAsrText(variant); + assert.ok(result.includes('一成系统'), `"${variant}" should correct to 一成系统 via aggressive regex, got "${result}"`); + }); + } +}); + +// ================================================================ +// 5. normalizeKnowledgeAlias —— 路由层额外归一化 +// ================================================================ +describe('normalizeKnowledgeAlias —— 路由层归一化', () => { + it('一成，系统（带标点间隔）→ 一成系统', () => { + const result = normalizeKnowledgeAlias('一成，系统'); + assert.ok(result.includes('一成系统'), `Got "${result}"`); + }); + + it('一成 系统（带空格间隔）→ 一成系统', () => { + const result = normalizeKnowledgeAlias('一成 系统'); + assert.ok(result.includes('一成系统'), `Got "${result}"`); + }); + + it('XX系统 → 一成系统', () => { + const result = normalizeKnowledgeAlias('XX系统'); + assert.ok(result.includes('一成系统'), `Got "${result}"`); + }); + + it('大窝 → 大沃', () => { + const result = normalizeKnowledgeAlias('大窝'); + assert.ok(result.includes('大沃'), `Got "${result}"`); + }); + + it('暖炉原理 → 火炉原理', () => { + const result = normalizeKnowledgeAlias('暖炉原理'); + assert.ok(result.includes('火炉原理'), `Got "${result}"`); + }); + + it('AI众享（大写）→ Ai众享', () => { + const result = normalizeKnowledgeAlias('AI众享怎么用'); + assert.ok(result.includes('Ai众享'), `Got "${result}"`); + }); + + it('圣咖学院 → 盛咖学愿', () => { + const result = normalizeKnowledgeAlias('圣咖学院'); + assert.ok(result.includes('盛咖学愿'), `Got "${result}"`); + }); +}); + +// ================================================================ +// 6. 完整ASR流水线 —— correctAsrText + normalizeKnowledgeAlias 组合 +// ================================================================ +describe('完整ASR流水线 —— 纠错+归一化组合', () => { + + describe('一成系统变体经流水线后应路由到KB', () => { + const pipelineCases = [ + '一城系统是什么', + '逸城系统怎么用', + '易成系统介绍', + '益生系统怎么样', + '义诚系统核心优势', + '壹成系统三大平台', + '一声系统有什么用', + '翼成系统赋能团队', + ]; + for (const raw of pipelineCases) { + it(`"${raw}" → 应路由到KB`, () => { + assertPipelineRouteKb(raw); + }); + } + }); + + describe('产品名ASR错误经流水线后应路由到KB', () => { + const productAsrCases = [ + ['小洪产品功效', '小红ASR错误'], + ['小宏怎么吃', '小红ASR错误'], + ['大百功效是什么', '大白ASR错误'], + ['大柏怎么吃', '大白ASR错误'], + ['小百怎么服用', '小白ASR错误'], + ['小柏功效', '小白ASR错误'], + ]; + for (const [raw, label] of productAsrCases) { + it(`${label}: "${raw}" → 应路由到KB`, () => { + assertPipelineRouteKb(raw); + }); + } + }); + + describe('其他ASR错误经流水线后应路由到KB', () => { + const otherAsrCases = [ + ['暖炉原理是什么意思', '暖炉→火炉'], + ['营养配送系统原理', '配送→保送'], + ['整应反应是什么', '整应→好转'], + ['排毒反应正常吗', '排毒反应→好转反应'], + ['盛咖学院怎么用', '学院→学愿'], + ['起步三观是什么', '三观→三关'], + ['double pm介绍', 'double pm→德国PM'], + ['阿玉吠陀是什么', '阿玉→阿育'], + ]; + for (const [raw, label] of otherAsrCases) { + it(`${label}: "${raw}" → 应路由到KB`, () => { + assertPipelineRouteKb(raw); + }); + } + }); +}); + +// ================================================================ +// 7. ASR识别失败/乱码/噪声场景 +// ================================================================ +describe('ASR识别失败/异常场景', () => { + + it('空字符串 → 不崩溃，返回空', () => { + assert.equal(correctAsrText(''), ''); + assert.equal(correctAsrText(null), ''); + assert.equal(correctAsrText(undefined), ''); + }); + + it('纯噪声标点 → 不崩溃', () => { + const result = correctAsrText('，，，。。。！！'); + assert.equal(typeof result, 'string'); + }); + + it('语气词噪声 → 不应路由到KB', () => { + const noises = ['嗯嗯嗯', '啊啊啊', '哦哦哦', '额额额']; + for (const noise of noises) { + const processed = fullAsrPipeline(noise); + const result = shouldForceKnowledgeRoute(processed); + assert.equal(result, false, `Noise "${noise}" should NOT route to KB`); + } + }); + + it('极短识别（单字/双字）→ 不应路由到KB', () => { + const shorts = ['嗯', '好', '啊', '是', '对', '哦']; + for (const s of shorts) { + const processed = fullAsrPipeline(s); + const result = shouldForceKnowledgeRoute(processed); + assert.equal(result, false, `Short "${s}" should NOT route to KB`); + } + }); + + it('混合中英文乱码 → 不崩溃，不误触发KB', () => { + const garbled = ['abc123你好', 'test test', '!!!???', '😊😊😊']; + for (const g of garbled) { + const processed = fullAsrPipeline(g); + assert.equal(typeof processed, 'string', 'Should return string'); + } + }); + + it('超长ASR文本 → 不崩溃', () => { + const longText = '我想问一下关于'.repeat(50) + '基础三合一怎么吃'; + const result = correctAsrText(longText); + assert.equal(typeof result, 'string'); + assert.ok(result.length > 0); + }); +}); + +// ================================================================ +// 8. ASR部分识别 —— 语音被截断的情况 +// ================================================================ +describe('ASR部分识别 —— 语音截断/不完整', () => { + + it('"一成系" → 不完整但不崩溃', () => { + const result = correctAsrText('一成系'); + assert.equal(typeof result, 'string'); + }); + + it('"基础三合" → 不完整，不应匹配PHRASE_MAP', () => { + const result = correctAsrText('基础三合'); + assert.ok(!result.includes('基础套装'), `Incomplete "基础三合" should not trigger full phrase mapping, got "${result}"`); + }); + + it('"小红怎" → 不完整但产品名应被扩展', () => { + const result = correctAsrText('小红怎'); + // 小红 后面是 怎，在suffix list里有 怎么 但没有单独的 怎 + assert.equal(typeof result, 'string'); + }); + + it('"德国P" → 不完整，不应触发', () => { + const result = correctAsrText('德国P'); + assert.ok(!result.includes('德国PM公司'), `Incomplete should not over-expand, got "${result}"`); + }); +}); + +// ================================================================ +// 9. 复合ASR错误 —— 一句话里有多个ASR错误 +// ================================================================ +describe('复合ASR错误 —— 一句话中包含多个识别错误', () => { + + it('"一城系统的小洪产品功效" → 一成系统 + 小红', () => { + const result = correctAsrText('一城系统的小洪产品功效'); + assert.ok(result.includes('一成系统'), `Should correct 一城→一成, got "${result}"`); + assert.ok(result.includes('小红'), `Should correct 小洪→小红, got "${result}"`); + }); + + it('"大百和小柏的区别" → 大白 + 小白', () => { + const result = correctAsrText('大百和小柏的区别'); + assert.ok(result.includes('大白'), `Should correct 大百→大白, got "${result}"`); + assert.ok(result.includes('小白'), `Should correct 小柏→小白, got "${result}"`); + }); + + it('"爱众享和盛咖学院" → Ai众享 + 盛咖学愿', () => { + const result = correctAsrText('爱众享和盛咖学院'); + assert.ok(result.includes('Ai众享'), `Should correct 爱众享→Ai众享, got "${result}"`); + assert.ok(result.includes('盛咖学愿'), `Should correct 盛咖学院→盛咖学愿, got "${result}"`); + }); + + it('"大窝的暖炉原理" → 大沃 + 火炉原理', () => { + const result = correctAsrText('大窝的暖炉原理'); + assert.ok(result.includes('大沃'), `Should correct 大窝→大沃, got "${result}"`); + assert.ok(result.includes('火炉原理'), `Should correct 暖炉→火炉, got "${result}"`); + }); + + it('复合错误经完整流水线后应路由到KB', () => { + assertPipelineRouteKb('一城系统的小洪怎么吃'); + assertPipelineRouteKb('大百和小柏的区别是什么'); + assertPipelineRouteKb('爱众享和盛咖学院介绍'); + }); +}); + +// ================================================================ +// 10. 真实语音场景模拟 —— 模拟用户真实说话方式的ASR输出 +// ================================================================ +describe('真实语音场景 —— 模拟实际用户说话的ASR识别结果', () => { + + it('"那个一城系统是干嘛的呀" → 应路由到KB', () => { + assertPipelineRouteKb('那个一城系统是干嘛的呀'); + }); + + it('"小洪和大百一起吃吗" → 应路由到KB', () => { + assertPipelineRouteKb('小洪和大百一起吃吗'); + }); + + it('"我想问一下那个暖炉原理是什么意思" → 应路由到KB', () => { + assertPipelineRouteKb('我想问一下那个暖炉原理是什么意思'); + }); + + it('"你们那个double pm是什么公司" → 应路由到KB', () => { + assertPipelineRouteKb('你们那个double pm是什么公司'); + }); + + it('"吃了以后有整应反应怎么办" → 应路由到KB', () => { + assertPipelineRouteKb('吃了以后有整应反应怎么办'); + }); + + it('"盛咖学院里面的课程怎么看" → 应路由到KB', () => { + assertPipelineRouteKb('盛咖学院里面的课程怎么看'); + }); + + it('"那个营养配送系统是怎么回事" → 应路由到KB', () => { + assertPipelineRouteKb('那个营养配送系统是怎么回事'); + }); + + it('"新人起步三观是什么" → 应路由到KB', () => { + assertPipelineRouteKb('新人起步三观是什么'); + }); + + it('"维适多怎么服用" → 应路由到KB', () => { + assertPipelineRouteKb('维适多怎么服用'); + }); + + it('"大窝能帮我介绍一下大百吗" → 应路由到KB', () => { + assertPipelineRouteKb('大窝能帮我介绍一下大百吗'); + }); + + it('"五加一活动是什么" → 应路由到KB', () => { + assertPipelineRouteKb('五加一活动是什么'); + }); + + it('"阿育费陀跟PM产品有什么关系" → 应路由到KB', () => { + assertPipelineRouteKb('阿育费陀跟PM产品有什么关系'); + }); +}); + +// ================================================================ +// 11. 负面用例 —— 正常文本不应被ASR纠错误改 +// ================================================================ +describe('负面用例 —— 正常文本不应被误纠', () => { + + it('"一成不变" → 不应被改为"一成系统不变"', () => { + const result = correctAsrText('一成不变'); + assert.ok(!result.includes('一成系统'), `"一成不变" should NOT be corrected, got "${result}"`); + }); + + it('"今天天气好" → 保持不变', () => { + const result = correctAsrText('今天天气好'); + assert.equal(result, '今天天气好'); + }); + + it('"你好" → 保持不变', () => { + const result = correctAsrText('你好'); + assert.equal(result, '你好'); + }); + + it('"谢谢你帮忙" → 保持不变', () => { + const result = correctAsrText('谢谢你帮忙'); + assert.equal(result, '谢谢你帮忙'); + }); + + it('"大白天出去" → 不应被纠正为PM产品', () => { + // "大白" 后面跟 "天" 不在suffix list，不应触发产品扩展 + const result = correctAsrText('大白天出去'); + assert.ok(!result.includes('Basics'), `"大白天出去" should NOT trigger product alias, got "${result}"`); + }); + + it('"小白兔" → 不应被纠正为PM产品', () => { + const result = correctAsrText('小白兔'); + assert.ok(!result.includes('Restorate'), `"小白兔" should NOT trigger product alias, got "${result}"`); + }); +}); + +console.log('\n=== ASR覆盖测试加载完成 ===\n'); diff --git a/test2/server/tests/test_kb_prompt_compare.js b/test2/server/tests/test_kb_prompt_compare.js new file mode 100644 index 0000000..d66f5cf --- /dev/null +++ b/test2/server/tests/test_kb_prompt_compare.js @@ -0,0 +1,241 @@ +/** + * KB Prompt A/B 对比测试 + * 用失败用例验证优化后的prompt效果 + * 零依赖：仅用Node内置模块 + */ +const path = require('path'); +const https = require('https'); +const fs = require('fs'); + +// 手动加载 .env +const envPath = path.join(__dirname, '../.env'); +if (fs.existsSync(envPath)) { + fs.readFileSync(envPath, 'utf8').split('\n').forEach(line => { + const trimmed = line.trim(); + if (!trimmed || trimmed.startsWith('#')) return; + const idx = trimmed.indexOf('='); + if (idx > 0) { + const key = trimmed.slice(0, idx).trim(); + let val = trimmed.slice(idx + 1).trim(); + if ((val.startsWith('"') && val.endsWith('"')) || (val.startsWith("'") && val.endsWith("'"))) { + val = val.slice(1, -1); + } + if (!process.env[key]) process.env[key] = val; + } + }); +} + +const kbHttpAgent = new https.Agent({ + keepAlive: true, + keepAliveMsecs: 30000, + maxSockets: 6, + timeout: 15000, +}); + +// ===== 旧prompt ===== +const OLD_PROMPT = '你是大沃的智能助手，负责回答与PM-FitLine德国产品相关的问题。产品用法、成分、剂量、价格等专业信息必须依据知识库，不得自行补充或猜测；公司背景、理念等常识性内容可适当补充。不得编造不存在的产品名称或数据。若知识库无相关内容，坦诚说明并建议查看产品说明书或咨询推荐人。回答口语化、简洁、专业，1-3句给结论，150字内。'; + +// ===== 新prompt ===== +const NEW_PROMPT = '你是大沃的智能助手，负责回答与PM-FitLine德国产品相关的问题。知识库涵盖近50款PM-FitLine产品的完整资料（成分、用法、剂量、价格、规格、搭配方案、好转反应等）及117个常见问答。回答规则：产品相关具体信息必须严格依据知识库，不得猜测或自行补充；公司背景、健康常识可适当补充。产品常有别名（小红=艾特维、大白=倍适、小白=维适多等），请注意识别。不得编造产品名或数据。PM是营养品非药物，涉及疾病建议咨询医生。若知识库无相关内容，坦诚说明并建议咨询推荐人。回答口语化、简洁，1-3句给结论，150字内。'; + +// ===== 全面测试用例：10个维度 ===== +const TEST_CASES = [ + // ── 1. 产品用法（营养品） ── + { cat: '用法', name: '细胞抗氧素怎么吃(原失败)', query: 'Apple Antioxy Zellschutz 细胞抗氧素 怎么吃', original: '细胞抗氧素怎么吃', expectHit: true }, + { cat: '用法', name: '小红怎么喝', query: 'Activize Oxyplus 艾特维 小红 怎么喝', original: '小红怎么喝', expectHit: true }, + { cat: '用法', name: '小白什么时候喝', query: 'Restorate 维适多 小白 什么时候喝 服用方法', original: '小白什么时候喝', expectHit: true }, + { cat: '用法', name: 'Q10怎么用', query: 'Q10辅酵素 怎么用 用法', original: 'Q10怎么用', expectHit: true }, + { cat: '用法', name: '心脏衰竭者小红用量', query: 'PM-Fitline 心脏衰竭 小红 用量 怎么喝', original: '心脏衰竭的人小红怎么喝', expectHit: true }, + + // ── 2. 产品价格 ── + { cat: '价格', name: '大白多少钱', query: 'Basics 倍适 大白 价格 多少钱', original: '大白多少钱', expectHit: true }, + { cat: '价格', name: '小绿排毒饮价格', query: 'D-Drink 小绿 排毒饮 价格 多少钱', original: '小绿多少钱', expectHit: true }, + { cat: '价格', name: '祛皱凝胶价格', query: 'Ultimate Young 三分钟瞬间祛皱凝胶 价格', original: '祛皱凝胶多少钱', expectHit: true }, + + // ── 3. 产品成分 ── + { cat: '成分', name: '小红里有什么成分', query: 'Activize Oxyplus 艾特维 小红 成分 配方', original: '小红里面有什么成分', expectHit: true }, + { cat: '成分', name: '美容饮成分', query: 'Beauty 肽美 美容饮 成分', original: '美容饮有什么成分', expectHit: true }, + + // ── 4. 产品搭配 ── + { cat: '搭配', name: '基础三合一是什么', query: '德国PM细胞营养素 基础套装 大白 小红 小白', original: '基础三合一是什么', expectHit: true }, + { cat: '搭配', name: '减肥搭配方案', query: 'TopShape ProShape 减肥 搭配 方案', original: '想减肥应该吃什么搭配', expectHit: true }, + + // ── 5. 好转反应 ── + { cat: '好转反应', name: '为什么会痒', query: 'PM-Fitline 好转反应 痒 发红', original: '喝了PM为什么会痒', expectHit: true }, + { cat: '好转反应', name: '喝了头晕怎么回事', query: 'PM-Fitline 好转反应 头晕 目眩', original: '喝完营养素头晕怎么回事', expectHit: true }, + { cat: '好转反应', name: '为什么喝小红胃痛', query: 'PM-Fitline 小红 艾特维 胃痛 好转反应', original: '为什么喝小红胃会痛', expectHit: true }, + + // ── 6. 别名识别 ── + { cat: '别名', name: '小黑是什么产品', query: 'MEN+ 倍力健 小黑 产品介绍', original: '小黑是什么', expectHit: true }, + { cat: '别名', name: '乐活50+适合谁', query: 'Generation 50+ 乐活50+ 适合谁 适用人群', original: '乐活50+适合谁吃', expectHit: true }, + + // ── 7. 护肤品 ── + { cat: '护肤', name: '眼霜怎么用', query: 'Eye Care 全效眼霜 用法 怎么用', original: '眼霜怎么用', expectHit: true }, + { cat: '护肤', name: '面膜多久敷一次', query: 'Hydrating-Shot Mask 靓白保湿面膜 多久 频率', original: '面膜多久敷一次', expectHit: true }, + + // ── 8. 117问答/疾病相关 ── + { cat: 'Q&A', name: '高血压能吃PM吗', query: 'PM-Fitline 高血压 能吃吗 影响', original: '高血压能吃PM吗', expectHit: true }, + { cat: 'Q&A', name: '糖尿病能减药吗', query: 'PM-Fitline 糖尿病 减药 停药', original: '吃PM后糖尿病的药能减吗', expectHit: true }, + { cat: 'Q&A', name: '小红咖啡因和咖啡一样吗', query: 'PM-Fitline 小红 艾特维 咖啡因 瓜拉纳 区别', original: '小红里的咖啡因跟咖啡一样吗', expectHit: true }, + + // ── 9. 边界测试（KB中不存在的信息） ── + { cat: '边界', name: '不存在的产品', query: 'PM-Fitline 超级修复胶囊', original: 'PM有超级修复胶囊这个产品吗', expectHit: false }, + { cat: '边界', name: '跟主题无关的问题', query: '今天天气怎么样', original: '今天天气怎么样', expectHit: false }, + + // ── 10. 公司/系统 ── + { cat: '公司', name: 'PM是什么公司', query: 'PM-International 德国PM公司 介绍', original: 'PM是什么公司', expectHit: true }, + + // ── 11. 一成系统 ── + { cat: '一成系统', name: '一成系统如何赋能德国PM', query: '一成系统 赋能 德国PM', original: '一成系统如何赋能德国PM', expectHit: true }, +]; + +function httpsPost(url, body, headers) { + return new Promise((resolve, reject) => { + const data = JSON.stringify(body); + const parsed = new URL(url); + const req = https.request({ + hostname: parsed.hostname, + path: parsed.pathname, + method: 'POST', + headers: { ...headers, 'Content-Length': Buffer.byteLength(data) }, + agent: kbHttpAgent, + timeout: 15000, + }, (res) => { + let chunks = []; + res.on('data', c => chunks.push(c)); + res.on('end', () => { + try { + resolve(JSON.parse(Buffer.concat(chunks).toString())); + } catch (e) { + reject(new Error('JSON parse error: ' + Buffer.concat(chunks).toString().slice(0, 200))); + } + }); + }); + req.on('error', reject); + req.on('timeout', () => { req.destroy(); reject(new Error('timeout')); }); + req.write(data); + req.end(); + }); +} + +async function callKB(systemPrompt, query, label) { + const endpointId = process.env.VOLC_ARK_KNOWLEDGE_ENDPOINT_ID || process.env.VOLC_ARK_ENDPOINT_ID; + const kbModel = process.env.VOLC_ARK_KB_MODEL || endpointId; + const authKey = process.env.VOLC_ARK_API_KEY || process.env.VOLC_ACCESS_KEY_ID; + const kbIds = (process.env.VOLC_ARK_KNOWLEDGE_BASE_IDS || '').split(',').map(id => id.trim()).filter(Boolean); + const topK = parseInt(process.env.VOLC_ARK_KNOWLEDGE_TOP_K) || 3; + const threshold = parseFloat(process.env.VOLC_ARK_KNOWLEDGE_THRESHOLD) || 0.3; + + const body = { + model: kbModel, + messages: [ + { role: 'system', content: systemPrompt }, + { role: 'user', content: query }, + ], + metadata: { + knowledge_base: { + dataset_ids: kbIds, + top_k: topK, + threshold: threshold, + }, + }, + stream: false, + max_tokens: 80, + thinking: { type: 'disabled' }, + }; + + const start = Date.now(); + try { + const res = await httpsPost( + 'https://ark.cn-beijing.volces.com/api/v3/chat/completions', + body, + { + 'Content-Type': 'application/json', + 'Authorization': `Bearer ${authKey}`, + } + ); + const elapsed = Date.now() - start; + const content = res?.choices?.[0]?.message?.content || '(empty)'; + const usage = res?.usage || {}; + const isNoHit = /未找到|未提及|暂无|没有相关|无法|不确定/.test(content); + return { label, content, elapsed, usage, hit: !isNoHit }; + } catch (err) { + return { label, content: `ERROR: ${err.message}`, elapsed: Date.now() - start, usage: {}, hit: false }; + } +} + +async function runTest() { + console.log('============================================================'); + console.log(' KB 新Prompt 全面对话测试（10维度 x 25用例）'); + console.log('============================================================\n'); + + const kbModel = process.env.VOLC_ARK_KB_MODEL || process.env.VOLC_ARK_KNOWLEDGE_ENDPOINT_ID || '(not set)'; + const threshold = process.env.VOLC_ARK_KNOWLEDGE_THRESHOLD || '0.3'; + console.log(`模型: ${kbModel} | 阈值: ${threshold} | 用例数: ${TEST_CASES.length}\n`); + + const results = []; + const catStats = {}; + + for (let i = 0; i < TEST_CASES.length; i++) { + const tc = TEST_CASES[i]; + console.log(`[${i + 1}/${TEST_CASES.length}] 【${tc.cat}】${tc.name}`); + console.log(` 原始问题: "${tc.original}"`); + + const r = await callKB(NEW_PROMPT, tc.query, tc.name); + const matchExpect = r.hit === tc.expectHit; + + // 对边界测试特殊处理：no-hit是期望结果 + const icon = matchExpect ? '✅' : '⚠️'; + const hitLabel = tc.expectHit + ? (r.hit ? 'HIT' : 'MISS(应命中)') + : (r.hit ? 'HIT(应未命中)' : 'NO-HIT(正确)'); + + console.log(` ${icon} ${hitLabel} | ${r.elapsed}ms | tokens: ${r.usage.prompt_tokens || '?'}→${r.usage.completion_tokens || '?'}`); + console.log(` 回答: ${r.content.slice(0, 150)}${r.content.length > 150 ? '...' : ''}\n`); + + results.push({ ...tc, ...r, matchExpect }); + + if (!catStats[tc.cat]) catStats[tc.cat] = { total: 0, correct: 0, totalMs: 0 }; + catStats[tc.cat].total++; + if (matchExpect) catStats[tc.cat].correct++; + catStats[tc.cat].totalMs += r.elapsed; + } + + // ── 分类汇总 ── + console.log('============================================================'); + console.log(' 分类汇总'); + console.log('============================================================'); + let totalCorrect = 0, totalMs = 0; + for (const [cat, s] of Object.entries(catStats)) { + const pct = Math.round(s.correct / s.total * 100); + const avgMs = Math.round(s.totalMs / s.total); + console.log(` 【${cat}】${s.correct}/${s.total} 正确 (${pct}%) | 平均 ${avgMs}ms`); + totalCorrect += s.correct; + totalMs += s.totalMs; + } + + // ── 总汇总 ── + console.log('\n============================================================'); + console.log(' 总汇总'); + console.log('============================================================'); + const totalPct = Math.round(totalCorrect / TEST_CASES.length * 100); + const avgMs = Math.round(totalMs / TEST_CASES.length); + console.log(` 总正确率: ${totalCorrect}/${TEST_CASES.length} (${totalPct}%)`); + console.log(` 平均延迟: ${avgMs}ms`); + + // ── 异常用例列表 ── + const issues = results.filter(r => !r.matchExpect); + if (issues.length > 0) { + console.log(`\n ⚠️ 异常用例 (${issues.length}个):`); + issues.forEach(r => { + console.log(` - 【${r.cat}】${r.name}: ${r.hit ? 'HIT' : 'NO-HIT'} (期望${r.expectHit ? 'HIT' : 'NO-HIT'})`); + }); + } else { + console.log('\n 🎉 全部用例符合预期！'); + } +} + +runTest().catch(err => { + console.error('测试失败:', err.message); + process.exit(1); +}); diff --git a/test2/server/tests/test_kb_protection.js b/test2/server/tests/test_kb_protection.js new file mode 100644 index 0000000..ffa5b08 --- /dev/null +++ b/test2/server/tests/test_kb_protection.js @@ -0,0 +1,541 @@ +/** + * KB保护窗口 + 质疑关键词 + query去噪 + 话题记忆 功能性测试 + * 针对"粉末"幻觉问题暴露的所有修复点进行回归测试 + * + * 运行方式: node --test tests/test_kb_protection.js + */ +const { describe, it } = require('node:test'); +const assert = require('node:assert/strict'); + +// ===== 被测模块加载 ===== +const { shouldForceKnowledgeRoute } = require('../services/realtimeDialogRouting'); +const { hasKnowledgeRouteKeyword } = require('../services/knowledgeKeywords'); +const contextKeywordTracker = require('../services/contextKeywordTracker'); + +// toolExecutor 需要特殊处理（有外部依赖），直接require静态方法 +let ToolExecutor; +try { + ToolExecutor = require('../services/toolExecutor'); +} catch (e) { + // toolExecutor可能依赖env，提供fallback + ToolExecutor = null; +} + +// ================================================================ +// 1. isKnowledgeFollowUp / shouldForceKnowledgeRoute 质疑场景测试 +// ================================================================ +describe('shouldForceKnowledgeRoute — 质疑/纠正类话术', () => { + // 为所有需要context的测试准备KB上下文 + const kbContext = [ + { role: 'user', content: '骨关节产品有哪些' }, + { role: 'assistant', content: '德国PM的健骨至尊氨糖软骨素胶囊...' }, + ]; + + // ---- 1. 直接否定 ---- + describe('直接否定', () => { + const cases = [ + '不是的', + '不是啊', + '不是不是', + '才不是', + '没有啊', + '没有吧', + '哪有', + '不是这么回事', + '不是这么说的', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true, + `"${text}" should be recognized as KB follow-up with KB context`); + }); + } + }); + + // ---- 2. 指出错误 ---- + describe('指出错误', () => { + const cases = [ + '你搞错了吧', + '说错了', + '弄错了', + '记错了', + '搞混了吧', + '你说反了', + '记岔了', + '张冠李戴', + '答非所问', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true, + `"${text}" should be recognized as KB follow-up`); + }); + } + }); + + // ---- 3. 说AI不对 ---- + describe('说AI不对', () => { + const cases = [ + '不对不对', + '你说的不对', + '不准确', + '说得不准', + '回答有误', + '不太对吧', + '说的有问题', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 4. 与自己认知矛盾 ---- + describe('与认知矛盾', () => { + const cases = [ + '跟我了解的不一样', + '我记得不是这样', + '我听说不是这样的', + '跟之前说的不一样', + '前后矛盾', + '你刚才不是说胶囊吗', + '别人告诉我是粉末', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 5. 怀疑/不信 ---- + describe('怀疑不信', () => { + const cases = [ + '我不信', + '骗人的吧', + '忽悠人呢', + '吹牛吧', + '太夸张了', + '离谱', + '扯淡', + '有依据吗', + '有证据吗', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 6. 要求复查 ---- + describe('要求复查', () => { + const cases = [ + '你再查查', + '再确认一下', + '重新查一下', + '核实一下', + '帮我再查一下', + '你查一下', + '搞清楚再说', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 7. 委婉质疑 ---- + describe('委婉质疑', () => { + const cases = [ + '好像不是这样吧', + '我觉得不太对', + '恐怕不是吧', + '感觉不对', + '我怎么记得不一样', + '印象中不是这样', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 8. 质问来源 ---- + describe('质问来源', () => { + const cases = [ + '谁说的', + '谁告诉你的', + '你从哪知道的', + '有什么根据', + '你确定吗', + '确定吗', + '真的吗', + '真的假的', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 9. 不可能/反问 ---- + describe('不可能/反问', () => { + const cases = [ + '怎么可能', + '不可能', + '不会吧', + '不是吧', + '开玩笑吧', + '别逗了', + '胡说', + '瞎说', + '别瞎说', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 10. 产品形态纠正 ---- + describe('产品形态纠正', () => { + const cases = [ + '粉末来的呀', + '是胶囊不是粉末', + '这个是冲剂', + '直接吞的', + '冲着喝的', + '是片剂', + '口服液来着', + '泡着喝', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 11. 纠正句式 ---- + describe('纠正句式', () => { + const cases = [ + '到底是粉末还是胶囊', + '究竟是什么形状', + '应该是冲剂吧', + '明明是粉末', + '其实是胶囊', + '怎么变成粉末了', + '不应该是这样', + ]; + for (const text of cases) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbContext), true); + }); + } + }); + + // ---- 复现"粉末"原始场景 ---- + describe('复现原始"粉末"幻觉场景', () => { + it('"粉末来的呀，你是搞错了吧？" → 必须走KB路由', () => { + assert.equal(shouldForceKnowledgeRoute('粉末来的呀，你是搞错了吧？', kbContext), true); + }); + + it('"粉末来的呀，你是搞错了吧？" → 无context也应命中（关键词直接匹配）', () => { + assert.equal(shouldForceKnowledgeRoute('粉末来的呀，你是搞错了吧？'), true); + }); + + it('"不对，是粉末不是胶囊" → 走KB路由', () => { + assert.equal(shouldForceKnowledgeRoute('不对，是粉末不是胶囊', kbContext), true); + }); + }); + + // ---- 负面用例：纯闲聊不应走KB ---- + describe('负面用例：纯闲聊不应触发KB', () => { + const cases = [ + '你好', + '谢谢', + '再见', + '今天天气好', + '哈哈哈', + '嗯嗯', + ]; + for (const text of cases) { + it(`"${text}" → 不应走KB路由`, () => { + assert.equal(shouldForceKnowledgeRoute(text), false, + `"${text}" should NOT route to KB`); + }); + } + }); +}); + +// ================================================================ +// 2. hasKnowledgeRouteKeyword — 新增质疑关键词命中测试 +// ================================================================ +describe('hasKnowledgeRouteKeyword — 新增质疑/产品形态关键词', () => { + describe('产品剂型直接命中', () => { + const keywords = ['粉末', '胶囊', '片剂', '冲剂', '口服液', '软胶囊', '颗粒', '膏状']; + for (const kw of keywords) { + it(`"${kw}" → 应直接命中KB关键词`, () => { + assert.equal(hasKnowledgeRouteKeyword(kw), true); + }); + } + }); + + describe('质疑词直接命中', () => { + const keywords = ['搞错了', '说错了', '不对', '确定吗', '真的吗', '不可能', '胡说', '离谱', '核实一下']; + for (const kw of keywords) { + it(`"${kw}" → 应直接命中KB关键词`, () => { + assert.equal(hasKnowledgeRouteKeyword(kw), true); + }); + } + }); + + describe('复合句子中的关键词命中', () => { + const cases = [ + '粉末来的呀你搞错了', + '这个产品不对，应该是胶囊', + '你确定吗这个是冲剂', + '谁说的这是片剂', + ]; + for (const text of cases) { + it(`"${text}" → 应命中KB关键词`, () => { + assert.equal(hasKnowledgeRouteKeyword(text), true); + }); + } + }); +}); + +// ================================================================ +// 3. sanitizeRewrittenQuery — 去噪截断测试 +// ================================================================ +describe('sanitizeRewrittenQuery — query后处理', () => { + // 只有ToolExecutor加载成功才测试 + const skip = !ToolExecutor || !ToolExecutor.sanitizeRewrittenQuery; + + it('清理口语填充词', { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery('骨关节啊怎么吃呢'); + assert.ok(!result.includes('啊'), `Should remove filler: got "${result}"`); + assert.ok(!result.includes('呢'), `Should remove filler: got "${result}"`); + assert.ok(result.includes('骨关节'), 'Should keep core content'); + assert.ok(result.includes('怎么吃'), 'Should keep core content'); + }); + + it('去除重复片段', { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery('骨关节 骨关节 怎么吃'); + const count = (result.match(/骨关节/g) || []).length; + assert.equal(count, 1, `Should dedupe: got "${result}"`); + }); + + it('截断超长query到80字符', { skip }, () => { + const longQuery = '德国PM细胞营养素基础套装大白小红小白一成系统NTC营养保送系统的功效作用成分配方怎么吃怎么用服用方法适合谁适用人群区别搭配原理价格多少钱哪里买'; + const result = ToolExecutor.sanitizeRewrittenQuery(longQuery); + assert.ok(result.length <= 80, `Should truncate to <=80: got len=${result.length}`); + }); + + it('空输入返回空', { skip }, () => { + assert.equal(ToolExecutor.sanitizeRewrittenQuery(''), ''); + assert.equal(ToolExecutor.sanitizeRewrittenQuery(null), ''); + }); + + it('正常query不应被破坏', { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery('德国PM骨关节产品怎么吃'); + assert.ok(result.includes('德国PM'), 'Should preserve content'); + assert.ok(result.includes('骨关节'), 'Should preserve content'); + assert.ok(result.includes('怎么吃'), 'Should preserve content'); + }); + + it('清理连续标点', { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery('骨关节？？！！怎么吃'); + assert.ok(!/[？！]{2,}/.test(result), `Should clean punctuation: got "${result}"`); + }); +}); + +// ================================================================ +// 4. enrichQueryWithContext — KB话题记忆优先级测试 +// ================================================================ +describe('enrichQueryWithContext — KB话题记忆优先', () => { + const sessionId = 'test_session_' + Date.now(); + + it('有KB话题记忆时，追问应关联KB话题而非历史keyword', () => { + // 模拟session有KB话题记忆 + const mockSession = { + _lastKbTopic: '骨关节产品有哪些', + _lastKbHitAt: Date.now(), + }; + + // 往keyword tracker中注入旧关键词（模拟上一个话题是"一成系统"） + contextKeywordTracker.updateSession(sessionId, '一成系统详细介绍'); + + const result = contextKeywordTracker.enrichQueryWithContext(sessionId, '怎么吃', mockSession); + + assert.ok(result.includes('骨关节'), `Should use KB topic memory: got "${result}"`); + assert.ok(!result.includes('一成系统'), `Should NOT use old keyword: got "${result}"`); + }); + + it('KB话题记忆过期后，回退到keyword tracker', () => { + const mockSession = { + _lastKbTopic: '骨关节产品有哪些', + _lastKbHitAt: Date.now() - 120000, // 2分钟前，已过期 + }; + + contextKeywordTracker.updateSession(sessionId, '一成系统详细介绍'); + const result = contextKeywordTracker.enrichQueryWithContext(sessionId, '怎么吃', mockSession); + + assert.ok(!result.includes('骨关节'), `Should NOT use expired KB topic: got "${result}"`); + }); + + it('无session时使用keyword tracker', () => { + contextKeywordTracker.updateSession(sessionId, '基础套装功效'); + const result = contextKeywordTracker.enrichQueryWithContext(sessionId, '怎么吃', null); + + assert.ok(result.includes('怎么吃'), `Should include query: got "${result}"`); + }); + + it('非追问类query不做enrichment', () => { + const mockSession = { + _lastKbTopic: '骨关节产品有哪些', + _lastKbHitAt: Date.now(), + }; + + const result = contextKeywordTracker.enrichQueryWithContext(sessionId, '德国PM公司介绍', mockSession); + assert.equal(result, '德国PM公司介绍', 'Non-follow-up should not be enriched'); + }); +}); + +// ================================================================ +// 5. KB保护窗口 — 集成场景模拟测试 +// ================================================================ +describe('KB保护窗口 — 场景模拟', () => { + // 模拟nativeVoiceGateway中的保护窗口逻辑 + function simulateProtectionWindow(cleanText, session) { + let isKnowledgeCandidate = shouldForceKnowledgeRoute(cleanText); + const KB_PROTECTION_WINDOW_MS = 60000; + if (!isKnowledgeCandidate && session._lastKbHitAt && (Date.now() - session._lastKbHitAt < KB_PROTECTION_WINDOW_MS)) { + const isPureChitchat = /^(喂|你好|嗨|谢谢|再见|拜拜|好的|嗯|哦|行|没事了|不用了|可以了)[，,。！？\s]*$/.test(cleanText); + if (!isPureChitchat) { + isKnowledgeCandidate = true; + } + } + return isKnowledgeCandidate; + } + + it('复现场景：KB hit后用户说"粉末来的呀你搞错了吧" → 应走KB', () => { + const session = { _lastKbHitAt: Date.now() - 5000, _lastKbTopic: '骨关节' }; + assert.equal(simulateProtectionWindow('粉末来的呀，你是搞错了吧？', session), true); + }); + + it('保护窗口内："你说的不对" → 应走KB', () => { + const session = { _lastKbHitAt: Date.now() - 10000 }; + assert.equal(simulateProtectionWindow('你说的不对', session), true); + }); + + it('保护窗口内："哦这样啊" → 非闲聊，应走KB', () => { + const session = { _lastKbHitAt: Date.now() - 10000 }; + assert.equal(simulateProtectionWindow('哦这样啊', session), true); + }); + + it('保护窗口内：纯闲聊"谢谢" → 不应走KB', () => { + const session = { _lastKbHitAt: Date.now() - 10000 }; + assert.equal(simulateProtectionWindow('谢谢', session), false); + }); + + it('保护窗口内：纯闲聊"好的" → 不应走KB', () => { + const session = { _lastKbHitAt: Date.now() - 10000 }; + assert.equal(simulateProtectionWindow('好的', session), false); + }); + + it('保护窗口内：纯闲聊"再见" → 不应走KB', () => { + const session = { _lastKbHitAt: Date.now() - 10000 }; + assert.equal(simulateProtectionWindow('再见', session), false); + }); + + it('保护窗口过期后："你搞错了" → 仍走KB（因为关键词直接命中）', () => { + const session = { _lastKbHitAt: Date.now() - 120000 }; + assert.equal(simulateProtectionWindow('你搞错了', session), true); + }); + + it('保护窗口过期后："天气怎么样" → 不走KB', () => { + const session = { _lastKbHitAt: Date.now() - 120000 }; + assert.equal(simulateProtectionWindow('今天心情好', session), false); + }); + + it('无KB历史时："随便聊聊" → 不走KB', () => { + const session = { _lastKbHitAt: 0 }; + assert.equal(simulateProtectionWindow('随便聊聊', session), false); + }); +}); + +// ================================================================ +// 6. 端到端场景回归测试（模拟完整对话链路） +// ================================================================ +describe('端到端场景回归 — 模拟"粉末"幻觉完整链路', () => { + it('场景1: 问骨关节→追问怎么吃→质疑粉末，三轮都应走KB', () => { + // 第1轮：问骨关节 + const r1 = shouldForceKnowledgeRoute('骨关节产品有哪些'); + assert.equal(r1, true, '第1轮"骨关节产品有哪些"应走KB'); + + // 第2轮：追问怎么吃（有上下文） + const ctx = [ + { role: 'user', content: '骨关节产品有哪些' }, + { role: 'assistant', content: '德国PM的健骨至尊氨糖软骨素胶囊...' }, + ]; + const r2 = shouldForceKnowledgeRoute('怎么吃', ctx); + assert.equal(r2, true, '第2轮"怎么吃"有KB上下文应走KB'); + + // 第3轮：质疑（关键词直接命中） + const r3 = shouldForceKnowledgeRoute('粉末来的呀，你是搞错了吧？'); + assert.equal(r3, true, '第3轮"粉末来的呀搞错了吧"应走KB'); + }); + + it('场景2: 用户用各种方式质疑，全部应走KB', () => { + const challengeTexts = [ + '粉末来的呀，你是搞错了吧？', + '不对不对，是粉末', + '你说的不对，这个是冲剂', + '我记得不是胶囊啊', + '你再查查，应该是粉末', + '好像不是这样吧', + '谁告诉你是胶囊的', + '怎么可能是胶囊', + '明明是粉末状的', + '冲着喝的不是吞的', + ]; + const ctx = [ + { role: 'user', content: '骨关节产品' }, + { role: 'assistant', content: '氨糖软骨素胶囊' }, + ]; + + for (const text of challengeTexts) { + const result = shouldForceKnowledgeRoute(text, ctx); + assert.equal(result, true, `质疑"${text}"应走KB路由`); + } + }); + + it('场景3: 正常追问也应走KB', () => { + const followUps = [ + '详细说说', + '怎么吃', + '多少钱', + '适合谁', + '功效是什么', + '成分是什么', + ]; + const ctx = [ + { role: 'user', content: '一成系统' }, + { role: 'assistant', content: '一成系统是德国PM的细胞营养素体系...' }, + ]; + + for (const text of followUps) { + const result = shouldForceKnowledgeRoute(text, ctx); + assert.equal(result, true, `追问"${text}"有KB上下文应走KB`); + } + }); +}); + +console.log('\n=== 测试文件加载完成，开始执行 ===\n'); diff --git a/test2/server/tests/test_kb_protection_extended.js b/test2/server/tests/test_kb_protection_extended.js new file mode 100644 index 0000000..90cd18e --- /dev/null +++ b/test2/server/tests/test_kb_protection_extended.js @@ -0,0 +1,576 @@ +/** + * KB保护窗口 + 质疑检测 + query去噪 + 话题记忆 深度扩展测试 + * 与 test_kb_protection.js 互补，覆盖更多边界、组合、时序场景 + * + * 运行方式: node --test tests/test_kb_protection_extended.js + */ +const { describe, it } = require('node:test'); +const assert = require('node:assert/strict'); + +const { shouldForceKnowledgeRoute, normalizeKnowledgeAlias } = require('../services/realtimeDialogRouting'); +const { hasKnowledgeRouteKeyword } = require('../services/knowledgeKeywords'); +const contextKeywordTracker = require('../services/contextKeywordTracker'); + +let ToolExecutor; +try { ToolExecutor = require('../services/toolExecutor'); } catch (e) { ToolExecutor = null; } + +const kbCtx = [ + { role: 'user', content: '基础三合一怎么吃' }, + { role: 'assistant', content: '大白早上空腹1平勺温水冲服，小红中午1平勺，小白睡前1平勺...' }, +]; + +// ================================================================ +// 1. shouldForceKnowledgeRoute — 组合质疑+产品名 +// ================================================================ +describe('组合质疑+产品名 —— 质疑词嵌入具体产品场景', () => { + const combos = [ + '大白不是这样吃的', + '小红功效你搞错了吧', + 'CC套装明明是乳霜', + '基础三合一不是冲剂', + 'Q10你说的不对', + 'D-Drink不是这么用的', + '一成系统跟我了解的不一样', + '火炉原理好像不是这么说的', + 'IB5不可能是这个功效吧', + '小白Restorate我记得不是这样', + '儿童倍适应该是胶囊不是粉末', + 'Hair+你再查查', + 'NTC你确定是这个原理吗', + '邓白氏谁说的AAA+', + '关节套装真的有这个功效吗', + 'TopShape太夸张了吧', + 'ProShape氨基酸骗人的吧', + '叶黄素我不信有这个作用', + '乳清蛋白说的有问题', + '运动饮料不是这个成分', + ]; + + for (const text of combos) { + it(`"${text}" → 应走KB`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbCtx), true, `"${text}" should route KB`); + }); + } +}); + +// ================================================================ +// 2. shouldForceKnowledgeRoute — 带标点/语气词的质疑变体 +// ================================================================ +describe('带标点/语气词的质疑变体', () => { + const variants = [ + '不对吧？', + '不对不对不对！', + '你搞错了吧！！', + '说错了，，，', + '我不信！真的假的？', + '骗人的吧……', + '太夸张了～', + '离谱啊！', + '扯淡吧？？', + '怎么可能？？？', + '不可能！不是吧！', + '好像不对哦～', + '你再查查？', + '核实一下嘛。', + '真的吗？真的吗？', + '谁说的啊？', + '有什么根据呢？', + '到底是什么啊！', + '应该是胶囊呀～', + '明明是粉末嘛！', + ]; + + for (const text of variants) { + it(`"${text}" → 应走KB`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbCtx), true, `"${text}" should route KB`); + }); + } +}); + +// ================================================================ +// 3. shouldForceKnowledgeRoute — 前缀剥离后的质疑 +// ================================================================ +describe('前缀剥离 —— 带"那/那你/你再/再"前缀的质疑', () => { + const prefixed = [ + '你再看看这个对不对', + '帮我再确认一下', + '你再看看吧', + '再来说说', + '麻烦你核实一下', + '帮我确认一下', + '那你确定吗', + '那再确认一下', + '那不对吧', + '那你搞错了', + '那我记得不是这样', + '再帮我查查', + '那再给我介绍一下', + '那详细说说', + '你再展开说说', + '那怎么吃', + '再讲讲功效是什么', + ]; + + for (const text of prefixed) { + it(`"${text}" → 应识别为KB追问`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbCtx), true, `"${text}" should route KB`); + }); + } +}); + +// ================================================================ +// 4. shouldForceKnowledgeRoute — 长句中嵌入质疑 +// ================================================================ +describe('长句中嵌入质疑 —— 质疑词不在句首', () => { + const longSentences = [ + '我刚才听你说的跟我了解的不一样', + '你之前的回答好像有误吧', + '按照我之前看到的资料应该是胶囊', + '怎么跟我之前在网上搜的不一致', + '别人告诉我是粉末的来着', + '但是我觉得你说的不太对', + '我看了很多资料你确定吗', + '感觉你说的和我了解的有出入', + '以前有人跟我说是冲着喝的', + '但是网上说法跟你不一样', + '我一直以为不是这样的', + '到底是什么意思啊', + '这些信息可靠吗有根据吗', + '我朋友说你讲的不对', + '这跟官方说的不一致吧', + ]; + + for (const text of longSentences) { + it(`"${text}" → 应走KB`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbCtx), true, `"${text}" should route KB`); + }); + } +}); + +// ================================================================ +// 5. shouldForceKnowledgeRoute — 纯追问模式（subject+action） +// ================================================================ +describe('subject+action追问模式 —— 带上下文', () => { + const subjectActions = [ + '这个怎么吃', + '那个功效是什么', + '它适合谁', + '这个产品多少钱', + '那个产品哪里买', + '这个怎么用', + '那个怎么操作', + '这个系统怎么配置', + '这个产品成分是什么', + '那个产品有什么功效', + '它怎么服用', + '这个有什么好处', + '那个配方', + '这个原理是什么', + '这个产品适合什么人', + '那个产品怎么买', + ]; + + for (const text of subjectActions) { + it(`"${text}" → 有上下文应走KB`, () => { + assert.equal(shouldForceKnowledgeRoute(text, kbCtx), true, `"${text}" should route KB with context`); + }); + } +}); + +// ================================================================ +// 6. shouldForceKnowledgeRoute — 否定用例扩展 +// ================================================================ +describe('否定用例扩展 —— 不应走KB的各类闲聊', () => { + const chitchat = [ + '你好', + '嗨', + '谢谢', + '再见', + '好的', + '嗯嗯', + '哈哈哈', + '拜拜', + '没事了', + '不用了', + '可以了', + '行', + '知道了', + '明白了', + '了解了', + '好吧', + '算了', + '今天天气好', + '你是谁', + '你叫什么名字', + '你是机器人吗', + '讲个笑话', + '唱首歌', + '几点了', + '我饿了', + '晚安', + '早上好', + '下午好', + '辛苦了', + '厉害', + ]; + + for (const text of chitchat) { + it(`"${text}" → 不应走KB`, () => { + assert.equal(shouldForceKnowledgeRoute(text), false, `"${text}" should NOT route KB`); + }); + } +}); + +// ================================================================ +// 7. hasKnowledgeRouteKeyword — 系统性全类别关键词覆盖 +// ================================================================ +describe('hasKnowledgeRouteKeyword — 产品名关键词系统覆盖', () => { + const productKeywords = [ + '大白', '小红', '小白', '基础三合一', 'Basics', 'Activize', 'Restorate', + '儿童倍适', 'CC套装', 'CC-Cell', 'Q10', 'IB5', 'D-Drink', + 'Hair+', 'ProShape氨基酸', 'Herbal Tea', 'TopShape', 'Men Face', + 'MEN+', '乐活', '草本茶', '叶黄素', '葡萄籽', '益生菌', + '胶原蛋白', '关节套装', '乳清蛋白', '运动饮料', '苹果细胞抗氧素', + ]; + + for (const kw of productKeywords) { + it(`产品"${kw}" → 应命中`, () => { + assert.equal(hasKnowledgeRouteKeyword(kw), true, `"${kw}" should match`); + }); + } +}); + +describe('hasKnowledgeRouteKeyword — FAQ/科学关键词系统覆盖', () => { + const faqKeywords = [ + '怎么吃', '功效', '成分', '多少钱', '价格', '适合谁', + '副作用', '多久见效', '见效', '好转反应', '是不是传销', + '传销', '是不是传销', '保质期', '哪里买', '怎么买', + 'NTC', '火炉原理', '阿育吠陀', '细胞营养素', + '正规吗', '合法吗', '贵不贵', '不舒服', + ]; + + for (const kw of faqKeywords) { + it(`FAQ"${kw}" → 应命中`, () => { + assert.equal(hasKnowledgeRouteKeyword(kw), true, `"${kw}" should match`); + }); + } +}); + +describe('hasKnowledgeRouteKeyword — 事业/培训关键词系统覆盖', () => { + const bizKeywords = [ + '招商', '代理', '加盟', '事业机会', '创业', '起步三关', + '精品会议', '成长上总裁', '做PM', '加入PM', 'PM事业', + ]; + + for (const kw of bizKeywords) { + it(`事业"${kw}" → 应命中`, () => { + assert.equal(hasKnowledgeRouteKeyword(kw), true, `"${kw}" should match`); + }); + } +}); + +describe('hasKnowledgeRouteKeyword — 质疑类关键词系统覆盖', () => { + const challengeKeywords = [ + '搞错了', '说错了', '弄错了', '不对', '不准确', '有误', + '确定吗', '真的吗', '不可能', '胡说', '骗人', '离谱', + '核实一下', '再查查', '粉末', '胶囊', '片剂', '冲剂', + '口服液', '软胶囊', '颗粒', '膏状', '到底是', '应该是', + '明明是', '不信', '吹牛', '扯淡', '有依据吗', '谁说的', + ]; + + for (const kw of challengeKeywords) { + it(`质疑"${kw}" → 应命中`, () => { + assert.equal(hasKnowledgeRouteKeyword(kw), true, `"${kw}" should match`); + }); + } +}); + +describe('hasKnowledgeRouteKeyword — 不应命中的普通词汇', () => { + const noMatch = [ + '你好', '天气', '笑话', '唱歌', '吃饭', '睡觉', + '电影', '音乐', '游戏', '旅游', '工作', '学习', + '开心', '难过', '累', '饿', '渴', '无聊', + ]; + + for (const kw of noMatch) { + it(`闲聊"${kw}" → 不应命中`, () => { + assert.equal(hasKnowledgeRouteKeyword(kw), false, `"${kw}" should NOT match`); + }); + } +}); + +// ================================================================ +// 8. sanitizeRewrittenQuery — 深度去噪测试 +// ================================================================ +describe('sanitizeRewrittenQuery — 深度去噪截断', () => { + const skip = !ToolExecutor || !ToolExecutor.sanitizeRewrittenQuery; + + const fillerCases = [ + ['骨关节啊嗯呢产品', '骨关节', '去除嗯啊呢'], + ['那个就是说这个呢功效是什么', '功效', '去除口语填充'], + ['骨关节哦嗯额功效', '骨关节', '去除多个语气词'], + ['基础三合一呀怎么吃呀', '基础三合一', '去除呀'], + ['嗯嗯那个小红功效', '小红', '去除嗯嗯那个'], + ]; + + for (const [input, expectContain, label] of fillerCases) { + it(`${label}: "${input}" → 含"${expectContain}"`, { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery(input); + assert.ok(result.includes(expectContain), `Got "${result}"`); + }); + } + + it('多次重复去重: "小红 小红 小红 功效"', { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery('小红 小红 小红 功效'); + const count = (result.match(/小红/g) || []).length; + assert.ok(count <= 2, `Should dedupe, got "${result}" (${count} occurrences)`); + }); + + it('去除连续空格', { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery('骨关节 产品 功效'); + assert.ok(!/ /.test(result), `Should remove multi-spaces, got "${result}"`); + }); + + const truncCases = [ + ''.padEnd(100, '德国PM细胞营养素基础套装大白小红小白'), + '这是一段超长的查询' + '关于产品的详细信息'.repeat(10), + ]; + + for (let i = 0; i < truncCases.length; i++) { + it(`超长截断 case ${i + 1}`, { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery(truncCases[i]); + assert.ok(result.length <= 80, `Should truncate, got len=${result.length}`); + }); + } + + it('特殊字符不崩溃', { skip }, () => { + const specials = ['骨关节\n产品', '基础三合一\t怎么吃', '小红\r\n功效']; + for (const s of specials) { + const result = ToolExecutor.sanitizeRewrittenQuery(s); + assert.equal(typeof result, 'string'); + } + }); + + it('已干净的query不被破坏', { skip }, () => { + const clean = '德国PM基础三合一 大白 小红 小白 怎么吃'; + const result = ToolExecutor.sanitizeRewrittenQuery(clean); + assert.ok(result.includes('基础三合一'), `Core preserved: got "${result}"`); + assert.ok(result.includes('怎么吃'), `Action preserved: got "${result}"`); + }); + + it('全标点输入', { skip }, () => { + const result = ToolExecutor.sanitizeRewrittenQuery('，，，。。。！！！'); + assert.equal(typeof result, 'string'); + }); +}); + +// ================================================================ +// 9. enrichQueryWithContext — 多场景深度测试 +// ================================================================ +describe('enrichQueryWithContext — 多场景深度', () => { + const sid = 'test_enrich_ext_' + Date.now(); + + it('新session空关键词 → 返回原始query', () => { + const result = contextKeywordTracker.enrichQueryWithContext('empty_sid_' + Date.now(), '怎么吃', null); + assert.ok(result.includes('怎么吃'), `Should return original: got "${result}"`); + }); + + it('有关键词+追问 → 关键词注入', () => { + const s = 'enrich_inject_' + Date.now(); + contextKeywordTracker.updateSession(s, '大白产品功效详细介绍'); + const result = contextKeywordTracker.enrichQueryWithContext(s, '怎么吃', null); + assert.ok(result.includes('怎么吃'), `Should include query: got "${result}"`); + }); + + it('非追问query → 不注入关键词', () => { + const s = 'enrich_noinject_' + Date.now(); + contextKeywordTracker.updateSession(s, '大白产品功效'); + const result = contextKeywordTracker.enrichQueryWithContext(s, '德国PM公司在哪里', null); + assert.equal(result, '德国PM公司在哪里', `Non-follow-up should not inject: got "${result}"`); + }); + + it('KB话题记忆优先于keyword tracker', () => { + const s = 'enrich_priority_' + Date.now(); + contextKeywordTracker.updateSession(s, '一成系统三大平台'); + const session = { _lastKbTopic: 'CC套装功效', _lastKbHitAt: Date.now() }; + const result = contextKeywordTracker.enrichQueryWithContext(s, '怎么吃', session); + assert.ok(result.includes('CC'), `Should use KB topic: got "${result}"`); + }); + + it('KB话题过期(>60s) → 降级到keyword tracker', () => { + const s = 'enrich_expired_' + Date.now(); + contextKeywordTracker.updateSession(s, '一成系统详细'); + const session = { _lastKbTopic: 'CC套装功效', _lastKbHitAt: Date.now() - 90000 }; + const result = contextKeywordTracker.enrichQueryWithContext(s, '怎么吃', session); + assert.ok(!result.includes('CC'), `Should NOT use expired KB topic: got "${result}"`); + }); + + it('多轮更新后取最近关键词', () => { + const s = 'enrich_multi_' + Date.now(); + contextKeywordTracker.updateSession(s, '大白产品功效'); + contextKeywordTracker.updateSession(s, '小红Activize怎么吃'); + contextKeywordTracker.updateSession(s, 'Q10辅酵素作用'); + const result = contextKeywordTracker.enrichQueryWithContext(s, '多少钱', null); + assert.ok(result.includes('多少钱'), `Should include query: got "${result}"`); + }); + + it('各种追问前缀都能触发enrichment', () => { + const s = 'enrich_prefixes_' + Date.now(); + contextKeywordTracker.updateSession(s, '大白产品功效'); + const followUps = ['怎么吃', '功效是什么', '多少钱', '适合谁', '成分是什么', '哪里买', '副作用', '什么意思', '怎么用', '他的规格是什么', '它的包装是什么', '这款是什么剂型', '那个是什么形态', '一天几次', '每天几次', '每日几次']; + for (const fup of followUps) { + const result = contextKeywordTracker.enrichQueryWithContext(s, fup, null); + assert.ok(result.includes(fup), `"${fup}" should be in result: got "${result}"`); + } + }); +}); + +// ================================================================ +// 10. KB保护窗口 — 精细时序测试 +// ================================================================ +describe('KB保护窗口 — 精细时序与边界', () => { + function simulateProtection(text, session) { + let isKb = shouldForceKnowledgeRoute(text); + const WINDOW = 60000; + if (!isKb && session._lastKbHitAt && (Date.now() - session._lastKbHitAt < WINDOW)) { + const isPureChitchat = /^(喂|你好|嗨|谢谢|再见|拜拜|好的|嗯|哦|行|没事了|不用了|可以了)[，,。！？\s]*$/.test(text); + if (!isPureChitchat) isKb = true; + } + return isKb; + } + + describe('窗口内（5s-55s）非闲聊提升', () => { + const timings = [5000, 10000, 20000, 30000, 45000, 55000, 59000, 59999]; + for (const t of timings) { + it(`${t}ms前KB hit + "哦这样啊" → 应走KB`, () => { + const session = { _lastKbHitAt: Date.now() - t }; + assert.equal(simulateProtection('哦这样啊', session), true); + }); + } + }); + + describe('窗口外（61s+）不提升', () => { + const timings = [60001, 65000, 120000, 300000]; + for (const t of timings) { + it(`${t}ms前KB hit + "哦这样啊" → 不走KB`, () => { + const session = { _lastKbHitAt: Date.now() - t }; + assert.equal(simulateProtection('哦这样啊', session), false); + }); + } + }); + + describe('窗口内各类纯闲聊不提升', () => { + const chitchat = ['你好', '嗨', '谢谢', '再见', '拜拜', '好的', '嗯', '哦', '行', '没事了', '不用了', '可以了', '喂']; + for (const c of chitchat) { + it(`窗口内"${c}" → 不走KB`, () => { + const session = { _lastKbHitAt: Date.now() - 5000 }; + assert.equal(simulateProtection(c, session), false); + }); + } + }); + + describe('窗口内各类非闲聊提升', () => { + const nonChat = [ + '然后呢', '还有吗', '继续', '还有什么', '那怎么办', + '这样可以吗', '有什么注意事项', '跟别的有什么区别', + '会不会有副作用', '我能吃吗', '孕妇可以吗', '小孩能吃吗', + '老人适合吗', '饭前还是饭后', '要吃多久', '一天几次', + ]; + for (const text of nonChat) { + it(`窗口内"${text}" → 应走KB`, () => { + const session = { _lastKbHitAt: Date.now() - 10000 }; + assert.equal(simulateProtection(text, session), true, `"${text}" should be elevated`); + }); + } + }); + + it('无KB历史(_lastKbHitAt=0) → 不提升', () => { + assert.equal(simulateProtection('然后呢', { _lastKbHitAt: 0 }), false); + }); + + it('无KB历史(undefined) → 不提升', () => { + assert.equal(simulateProtection('然后呢', {}), false); + }); + + it('_lastKbHitAt=null → 不提升', () => { + assert.equal(simulateProtection('然后呢', { _lastKbHitAt: null }), false); + }); +}); + +// ================================================================ +// 11. 端到端多轮模拟 — 更多变体场景 +// ================================================================ +describe('端到端多轮模拟 — 更多变体', () => { + + it('3轮：产品→追问→质疑价格', () => { + assert.equal(shouldForceKnowledgeRoute('Q10辅酵素功效'), true); + const ctx = [{ role: 'user', content: 'Q10辅酵素功效' }, { role: 'assistant', content: 'Q10...' }]; + assert.equal(shouldForceKnowledgeRoute('多少钱', ctx), true); + assert.equal(shouldForceKnowledgeRoute('太贵了吧，你确定吗'), true); + }); + + it('3轮：公司→认证→怀疑合法性', () => { + assert.equal(shouldForceKnowledgeRoute('德国PM公司介绍'), true); + assert.equal(shouldForceKnowledgeRoute('邓白氏AAA+认证'), true); + const ctx = [{ role: 'user', content: '邓白氏' }, { role: 'assistant', content: '邓白氏是...' }]; + assert.equal(shouldForceKnowledgeRoute('我不信，网上说是传销', ctx), true); + }); + + it('4轮：系统→功能→质疑→再查', () => { + assert.equal(shouldForceKnowledgeRoute('一成系统介绍'), true); + const ctx1 = [{ role: 'user', content: '一成系统' }, { role: 'assistant', content: '一成系统...' }]; + assert.equal(shouldForceKnowledgeRoute('行动圈是什么', ctx1), true); + assert.equal(shouldForceKnowledgeRoute('跟我了解的不一样'), true); + assert.equal(shouldForceKnowledgeRoute('你再查查一成系统'), true); + }); + + it('5轮：产品A→产品B→对比→质疑→纠正', () => { + assert.equal(shouldForceKnowledgeRoute('大白怎么吃'), true); + assert.equal(shouldForceKnowledgeRoute('小红怎么吃'), true); + assert.equal(shouldForceKnowledgeRoute('大白和小红有什么区别'), true); + const ctx = [{ role: 'user', content: '区别' }, { role: 'assistant', content: '大白是基础...' }]; + assert.equal(shouldForceKnowledgeRoute('你搞混了吧', ctx), true); + assert.equal(shouldForceKnowledgeRoute('应该是小红提供能量大白补充矿物质'), true); + }); + + it('连续4次质疑不同方式', () => { + assert.equal(shouldForceKnowledgeRoute('小白功效'), true); + const ctx = [{ role: 'user', content: '小白功效' }, { role: 'assistant', content: '小白...' }]; + assert.equal(shouldForceKnowledgeRoute('不对吧', ctx), true); + assert.equal(shouldForceKnowledgeRoute('你再查查', ctx), true); + assert.equal(shouldForceKnowledgeRoute('我不信', ctx), true); + assert.equal(shouldForceKnowledgeRoute('有什么根据', ctx), true); + }); + + it('KB话题→闲聊打断→再回到KB话题', () => { + assert.equal(shouldForceKnowledgeRoute('CC套装怎么用'), true); + assert.equal(shouldForceKnowledgeRoute('谢谢'), false); + assert.equal(shouldForceKnowledgeRoute('CC套装适合谁'), true); + }); +}); + +// ================================================================ +// 12. normalizeKnowledgeAlias — 更多归一化场景 +// ================================================================ +describe('normalizeKnowledgeAlias — 更多归一化场景', () => { + const cases = [ + ['一成，，系统', '一成系统', '多标点分隔'], + ['一成、系统', '一成系统', '顿号分隔'], + ['一成 系统', '一成系统', '多空格分隔'], + ['大我产品', '大沃', '大我→大沃'], + ['大卧介绍', '大沃', '大卧→大沃'], + ['哎众享怎么用', 'Ai众享', '哎众享→Ai众享'], + ['艾众享是什么', 'Ai众享', '艾众享→Ai众享'], + ['盛卡学愿介绍', '盛咖学愿', '盛卡→盛咖'], + ['圣咖学院怎么用', '盛咖学愿', '圣咖学院→盛咖学愿'], + ]; + + for (const [input, expectContain, label] of cases) { + it(`${label}: "${input}" → 含"${expectContain}"`, () => { + const result = normalizeKnowledgeAlias(input); + assert.ok(result.includes(expectContain), `Got "${result}"`); + }); + } +}); + +console.log('\n=== KB保护扩展测试加载完成 ===\n'); diff --git a/test2/server/tests/test_kb_scenarios.js b/test2/server/tests/test_kb_scenarios.js new file mode 100644 index 0000000..bf9e231 --- /dev/null +++ b/test2/server/tests/test_kb_scenarios.js @@ -0,0 +1,593 @@ +/** + * 基于知识库实际内容的功能性测试 + * 覆盖：单KB查询、多KB查询（话题切换）、追问+质疑混合、确定性改写、热答案匹配 + * + * 运行方式: node --test tests/test_kb_scenarios.js + */ +const { describe, it } = require('node:test'); +const assert = require('node:assert/strict'); + +const { shouldForceKnowledgeRoute } = require('../services/realtimeDialogRouting'); +const { hasKnowledgeRouteKeyword } = require('../services/knowledgeKeywords'); +const ToolExecutor = require('../services/toolExecutor'); + +// ================================================================ +// 辅助函数 +// ================================================================ +function assertKbRoute(text, ctx, msg) { + assert.equal(shouldForceKnowledgeRoute(text, ctx), true, msg || `"${text}" should route to KB`); +} +function assertNotKbRoute(text, ctx, msg) { + assert.equal(shouldForceKnowledgeRoute(text, ctx), false, msg || `"${text}" should NOT route to KB`); +} +function buildCtx(pairs) { + return pairs.map(([role, content]) => ({ role, content })); +} + +// ================================================================ +// 1. 单知识库查询 —— 每个产品/话题独立首轮查询 +// ================================================================ +describe('单KB查询 —— 产品类', () => { + const productQueries = [ + ['基础三合一怎么吃', '基础三合一'], + ['大白产品有什么功效', '大白Basics'], + ['小红Activize的作用是什么', '小红Activize'], + ['小白Restorate怎么服用', '小白Restorate'], + ['儿童倍适适合几岁的孩子', '儿童倍适'], + ['CC套装怎么用', 'CC套装'], + ['Q10辅酵素有什么功效', 'Q10辅酵素'], + ['IB5口腔喷雾怎么用', 'IB5口腔喷雾'], + ['D-Drink小绿排毒饮怎么用', 'D-Drink'], + ['Hair+发宝怎么用', 'Hair+发宝'], + ['运动饮料Fitness-Drink是什么', 'Fitness-Drink'], + ['TopShape纤萃减肥产品', 'TopShape'], + ['Generation 50+乐活产品', 'Generation 50+'], + ['Apple Antioxy细胞抗氧素功效', 'Apple Antioxy'], + ['ProShape氨基酸BCAA是什么', 'ProShape'], + ['Herbal Tea草本茶功效', 'Herbal Tea'], + ['Med Dental+草本护理牙膏', 'Med Dental+'], + ['Men Face男士护肤乳霜', 'Men Face'], + ['叶黄素产品怎么吃', '叶黄素'], + ['关节套装关节舒缓怎么用', '关节套装'], + ['乳清蛋白粉适合谁', '乳清蛋白'], + ['乐活奶昔怎么喝', '乐活奶昔'], + ]; + + for (const [query, label] of productQueries) { + it(`${label}: "${query}" → 应走KB路由`, () => { + assertKbRoute(query); + }); + } +}); + +describe('单KB查询 —— 系统/平台类', () => { + const systemQueries = [ + ['一成系统是什么', '一成系统'], + ['三大平台介绍一下', '三大平台'], + ['四大AI生态是什么', '四大AI生态'], + ['行动圈怎么用', '行动圈'], + ['盟主社区是什么', '盟主社区'], + ['AI众享是什么', 'AI众享'], + ['数字化工作室怎么用', '数字化工作室'], + ['盛咖学愿培训平台', '盛咖学愿'], + ]; + + for (const [query, label] of systemQueries) { + it(`${label}: "${query}" → 应走KB路由`, () => { + assertKbRoute(query); + }); + } +}); + +describe('单KB查询 —— 科学原理类', () => { + const scienceQueries = [ + ['NTC营养保送系统是什么原理', 'NTC'], + ['火炉原理是什么意思', '火炉原理'], + ['阿育吠陀是什么', '阿育吠陀'], + ]; + + for (const [query, label] of scienceQueries) { + it(`${label}: "${query}" → 应走KB路由`, () => { + assertKbRoute(query); + }); + } +}); + +describe('单KB查询 —— 公司/认证类', () => { + const companyQueries = [ + ['德国PM公司介绍', '德国PM'], + ['PM公司地址和电话', '地址电话'], + ['邓白氏认证是什么', '邓白氏'], + ['DSN全球100强', 'DSN'], + ['ELAB科隆名单认证', 'ELAB'], + ['PM是不是传销', '合法性'], + ['Rolf Sorg是谁', '创始人'], + ['宣明会慈善合作', '宣明会'], + ['培安烟台工厂', '培安'], + ]; + + for (const [query, label] of companyQueries) { + it(`${label}: "${query}" → 应走KB路由`, () => { + assertKbRoute(query); + }); + } +}); + +describe('单KB查询 —— FAQ常见问题类', () => { + const faqQueries = [ + ['PM产品多久见效', '见效时间'], + ['好转反应是什么', '好转反应'], + ['为什么要全套搭配使用', '全套搭配'], + ['和其他保健品有什么区别', '保健品区别'], + ['孕妇能吃PM产品吗', '特殊人群'], + ['产品多少钱', '价格'], + ['怎么加入PM', '加入方式'], + ['PM产品能治病吗', '治病声明'], + ]; + + for (const [query, label] of faqQueries) { + it(`${label}: "${query}" → 应走KB路由`, () => { + assertKbRoute(query); + }); + } +}); + +describe('单KB查询 —— 事业发展类', () => { + const businessQueries = [ + ['如何发展PM事业', '事业发展'], + ['线上拓客怎么做', '线上拓客'], + ['招商代理政策', '招商'], + ['新人起步三关是什么', '新人培训'], + ['为什么选择德国PM', '选择理由'], + ['陌生客户怎么沟通PM事业', '陌生沟通'], + ]; + + for (const [query, label] of businessQueries) { + it(`${label}: "${query}" → 应走KB路由`, () => { + assertKbRoute(query); + }); + } +}); + +// ================================================================ +// 2. 多KB查询 —— 话题切换场景 +// ================================================================ +describe('多KB查询 —— 话题切换（同一会话中切换不同产品/话题）', () => { + + it('场景1: 大白→小红→小白 三个产品连续查询', () => { + assertKbRoute('大白产品功效是什么'); + + const ctx1 = buildCtx([['user', '大白产品功效是什么'], ['assistant', '德国PM大白Basics是基础营养素...']]); + assertKbRoute('那小红呢', ctx1); + + const ctx2 = buildCtx([ + ['user', '大白产品功效是什么'], ['assistant', '德国PM大白Basics...'], + ['user', '那小红呢'], ['assistant', 'FitLine小红Activize...'], + ]); + assertKbRoute('小白怎么吃', ctx2); + }); + + it('场景2: 产品→公司→再回到产品', () => { + assertKbRoute('基础三合一介绍一下'); + + assertKbRoute('德国PM公司是什么时候成立的'); + + const ctx = buildCtx([ + ['user', '德国PM公司介绍'], ['assistant', '德国PM-International是1993年创立的...'], + ]); + assertKbRoute('那他们的产品有哪些', ctx); + }); + + it('场景3: 产品→一成系统→培训', () => { + assertKbRoute('CC套装怎么用'); + assertKbRoute('一成系统是什么'); + assertKbRoute('新人起步三关怎么做'); + }); + + it('场景4: 科学原理→产品→FAQ', () => { + assertKbRoute('NTC营养保送系统原理'); + assertKbRoute('大白Basics功效'); + assertKbRoute('多久能见效'); + }); + + it('场景5: 合法性→公司→产品→事业', () => { + assertKbRoute('PM是不是传销'); + assertKbRoute('邓白氏AAA+是什么'); + const ctx5 = buildCtx([['user', '邓白氏AAA+'], ['assistant', '邓白氏是全球最权威的商业信用评估机构...']]); + assertKbRoute('那产品有哪些', ctx5); + assertKbRoute('怎么加入PM事业'); + }); + + it('场景6: 快速切换5个不同产品', () => { + const products = [ + 'D-Drink排毒饮怎么用', + 'Q10辅酵素功效', + 'IB5口腔喷雾是什么', + '叶黄素怎么吃', + '关节套装适合谁', + ]; + for (const q of products) { + assertKbRoute(q); + } + }); +}); + +// ================================================================ +// 3. 追问场景（有上下文） +// ================================================================ +describe('追问场景 —— 基于上下文的知识库追问', () => { + + it('聊基础三合一后追问"怎么吃"', () => { + const ctx = buildCtx([['user', '基础三合一介绍'], ['assistant', '基础三合一包含大白小红小白...']]); + assertKbRoute('怎么吃', ctx); + }); + + it('聊小红后追问"功效是什么"', () => { + const ctx = buildCtx([['user', '小红产品'], ['assistant', 'FitLine小红Activize Oxyplus...']]); + assertKbRoute('功效是什么', ctx); + }); + + it('聊一成系统后追问"怎么用"', () => { + const ctx = buildCtx([['user', '一成系统介绍'], ['assistant', '一成系统是德国PM事业...']]); + assertKbRoute('怎么用', ctx); + }); + + it('聊CC套装后追问"适合谁"', () => { + const ctx = buildCtx([['user', 'CC套装功效'], ['assistant', 'CC套装含有葡萄籽提取物...']]); + assertKbRoute('适合谁', ctx); + }); + + it('聊火炉原理后追问"什么意思"', () => { + const ctx = buildCtx([['user', '火炉原理'], ['assistant', '火炉原理是PM产品的核心理念...']]); + assertKbRoute('什么意思', ctx); + }); + + it('聊D-Drink后追问"多少钱"', () => { + const ctx = buildCtx([['user', 'D-Drink小绿怎么用'], ['assistant', 'D-Drink小绿是14天排毒饮料...']]); + assertKbRoute('多少钱', ctx); + }); + + it('聊邓白氏后追问"什么意思"', () => { + const ctx = buildCtx([['user', '邓白氏AAA+认证'], ['assistant', '邓白氏是全球最权威的...']]); + assertKbRoute('什么意思', ctx); + }); + + it('聊NTC后追问"有什么好处"', () => { + const ctx = buildCtx([['user', 'NTC营养保送系统'], ['assistant', 'NTC营养保送系统是PM的核心技术...']]); + assertKbRoute('有什么好处', ctx); + }); + + it('聊Hair+后追问"成分是什么"', () => { + const ctx = buildCtx([['user', 'Hair+发宝怎么用'], ['assistant', 'Hair+包含口服发宝和外用发健...']]); + assertKbRoute('成分是什么', ctx); + }); + + it('用代词追问"这个产品怎么吃"', () => { + const ctx = buildCtx([['user', '小白Restorate功效'], ['assistant', '德国PM小白Restorate的核心功效是夜间修复...']]); + assertKbRoute('这个产品怎么吃', ctx); + }); +}); + +// ================================================================ +// 4. KB查询 + 质疑混合场景 +// ================================================================ +describe('KB查询+质疑混合 —— 用户查询后对回答产生质疑', () => { + + it('场景1: 问基础三合一→AI说冲剂→用户纠正"不对，是胶囊"', () => { + assertKbRoute('基础三合一怎么吃'); + const ctx = buildCtx([['user', '基础三合一怎么吃'], ['assistant', '基础三合一这样吃...']]); + assertKbRoute('不对，是胶囊不是冲剂', ctx); + }); + + it('场景2: 问小红功效→用户质疑"你搞错了吧"', () => { + assertKbRoute('小红产品功效'); + const ctx = buildCtx([['user', '小红产品功效'], ['assistant', '小红Activize...']]); + assertKbRoute('你搞错了吧', ctx); + }); + + it('场景3: 问NTC原理→用户说"我听说不是这样的"', () => { + assertKbRoute('NTC营养保送系统原理'); + const ctx = buildCtx([['user', 'NTC原理'], ['assistant', 'NTC营养保送系统...']]); + assertKbRoute('我听说不是这样的', ctx); + }); + + it('场景4: 问传销→用户要求"再查一下"', () => { + assertKbRoute('PM是不是传销'); + const ctx = buildCtx([['user', 'PM是不是传销'], ['assistant', '德国PM不是传销...']]); + assertKbRoute('你再查查，我看网上说法不一样', ctx); + }); + + it('场景5: 问价格→用户质疑"太贵了，你确定吗"', () => { + assertKbRoute('产品多少钱'); + const ctx = buildCtx([['user', '产品多少钱'], ['assistant', '产品价格因国家和地区有所不同...']]); + assertKbRoute('你确定吗，怎么那么贵', ctx); + }); + + it('场景6: 问CC套装→用户说"明明是乳霜不是胶囊"', () => { + assertKbRoute('CC套装是什么'); + const ctx = buildCtx([['user', 'CC套装'], ['assistant', 'CC套装包含CC-Cell胶囊和乳霜...']]); + assertKbRoute('明明是乳霜不是胶囊', ctx); + }); + + it('场景7: 问好转反应→用户说"骗人的吧"', () => { + assertKbRoute('好转反应是怎么回事'); + const ctx = buildCtx([['user', '好转反应'], ['assistant', '这是正常的好转反应...']]); + assertKbRoute('骗人的吧，有科学依据吗', ctx); + }); + + it('场景8: 问一成系统→用户说"跟我了解的不一样"', () => { + assertKbRoute('一成系统介绍'); + const ctx = buildCtx([['user', '一成系统'], ['assistant', '一成系统是德国PM事业发展的智能赋能工具...']]); + assertKbRoute('跟我了解的不一样啊', ctx); + }); + + it('场景9: 问D-Drink→用户说"这个是泡着喝的吧"', () => { + assertKbRoute('D-Drink怎么用'); + const ctx = buildCtx([['user', 'D-Drink怎么用'], ['assistant', 'D-Drink小绿是14天排毒饮料...']]); + assertKbRoute('这个是泡着喝的吧', ctx); + }); + + it('场景10: 问邓白氏→用户说"谁说的？有什么根据"', () => { + assertKbRoute('邓白氏评级是什么'); + const ctx = buildCtx([['user', '邓白氏'], ['assistant', '邓白氏是全球最权威的商业信用评估机构...']]); + assertKbRoute('谁说的？有什么根据', ctx); + }); +}); + +// ================================================================ +// 5. 多轮复杂对话场景（查询→追问→质疑→再查→切换话题） +// ================================================================ +describe('多轮复杂对话 —— 模拟真实用户完整会话', () => { + + it('5轮对话: 产品查询→追问→质疑→纠正→切换话题', () => { + // 轮1: 直接问产品 + assertKbRoute('基础三合一是什么'); + + // 轮2: 追问怎么吃 + const ctx2 = buildCtx([['user', '基础三合一是什么'], ['assistant', '基础三合一包含大白小红小白...']]); + assertKbRoute('怎么吃', ctx2); + + // 轮3: 质疑回答 + const ctx3 = buildCtx([ + ['user', '基础三合一是什么'], ['assistant', '基础三合一包含大白小红小白...'], + ['user', '怎么吃'], ['assistant', '大白早上空腹1平勺...'], + ]); + assertKbRoute('你说的温度不对吧', ctx3); + + // 轮4: 用户纠正 + assertKbRoute('应该是40度以下的水', ctx3); + + // 轮5: 切换到完全不同的话题 + assertKbRoute('PM是不是传销'); + }); + + it('6轮对话: 系统→子功能→质疑→公司→产品→FAQ', () => { + assertKbRoute('一成系统介绍'); + + const ctx2 = buildCtx([['user', '一成系统'], ['assistant', '一成系统是德国PM事业发展的智能赋能工具...']]); + assertKbRoute('行动圈是什么', ctx2); + + const ctx3 = buildCtx([['user', '一成系统'], ['assistant', '一成系统是德国PM事业发展的智能赋能工具...'], ['user', '行动圈是什么'], ['assistant', '行动圈是数字化工作室里的团队管理功能...']]); + assertKbRoute('好像不是这样吧', ctx3); + + assertKbRoute('德国PM公司背景'); + assertKbRoute('大白产品功效'); + assertKbRoute('孕妇能吃吗'); + }); + + it('4轮对话: 连续质疑同一个话题', () => { + assertKbRoute('火炉原理是什么'); + + const ctx1 = buildCtx([['user', '火炉原理'], ['assistant', '火炉原理是PM产品的核心理念比喻...']]); + assertKbRoute('不对，我记得不是这么说的', ctx1); + assertKbRoute('你再查查，应该是另一种说法', ctx1); + assertKbRoute('算了，到底是什么意思', ctx1); + }); +}); + +// ================================================================ +// 6. 确定性改写验证 +// ================================================================ +describe('确定性改写 —— buildDeterministicKnowledgeQuery', () => { + + describe('产品直接改写', () => { + const cases = [ + ['基础三合一怎么吃', '基础套装'], + ['大白产品', 'Basics'], + ['小红Activize功效', 'Activize'], + ['小白Restorate成分', 'Restorate'], + ['儿童倍适怎么吃', '儿童倍适'], + ['CC套装功效', 'CC'], + ['Q10辅酵素作用', 'Q10'], + ['IB5口腔喷雾', 'IB5'], + ['D-Drink排毒', 'D-Drink'], + ]; + + for (const [query, expectContain] of cases) { + it(`"${query}" → 改写应含"${expectContain}"`, () => { + const result = ToolExecutor.buildDeterministicKnowledgeQuery(query, []); + assert.ok(result, `Should have deterministic rewrite for "${query}"`); + assert.ok(result.includes(expectContain), + `Rewrite should contain "${expectContain}", got "${result}"`); + }); + } + }); + + describe('上下文追问改写', () => { + it('上下文有"大白"时追问"怎么吃" → 改写含Basics', () => { + const ctx = [{ role: 'assistant', content: '大白Basics是基础营养素...' }]; + const result = ToolExecutor.buildDeterministicKnowledgeQuery('怎么吃', ctx); + assert.ok(result, 'Should rewrite'); + assert.ok(result.includes('Basics'), `Should contain Basics, got "${result}"`); + }); + + it('上下文有"一成系统"时追问"怎么用" → 改写含一成系统', () => { + const ctx = [{ role: 'assistant', content: '一成系统是德国PM事业发展...' }]; + const result = ToolExecutor.buildDeterministicKnowledgeQuery('怎么用', ctx); + assert.ok(result, 'Should rewrite'); + assert.ok(result.includes('一成系统'), `Should contain 一成系统, got "${result}"`); + }); + + it('上下文有"火炉原理"时追问"什么意思" → 改写含火炉原理', () => { + const ctx = [{ role: 'assistant', content: '火炉原理是PM产品的核心理念...' }]; + const result = ToolExecutor.buildDeterministicKnowledgeQuery('什么意思', ctx); + assert.ok(result === '火炉原理', `Should rewrite to 火炉原理, got "${result}"`); + }); + }); + + describe('无匹配时返回空', () => { + it('"今天天气好" → 无确定性改写', () => { + const result = ToolExecutor.buildDeterministicKnowledgeQuery('今天天气好', []); + assert.equal(result, '', 'Chitchat should not have deterministic rewrite'); + }); + }); +}); + +// ================================================================ +// 7. 热答案匹配测试 +// ================================================================ +describe('热答案匹配 —— matchHotAnswer 验证', () => { + // matchHotAnswer是模块内部函数，通过ToolExecutor.execute间接调用 + // 这里直接测试确定性改写+KB路由来验证热答案可达性 + + const hotTopics = [ + ['基础三合一怎么吃', '基础三合一吃法'], + ['PM是不是传销', '合法性'], + ['NTC核心优势是什么', 'NTC核心优势'], + ['多久见效', '见效时间'], + ['为什么要全套搭配', '全套搭配原因'], + ['好转反应是什么', '好转反应'], + ['德国PM公司介绍', '公司介绍'], + ['小红功效', '小红功效'], + ['大白功效', '大白功效'], + ['小白功效', '小白功效'], + ['和其他保健品有什么区别', '保健品区别'], + ['CC套装怎么用', 'CC套装'], + ['Q10功效', 'Q10功效'], + ['IB5怎么用', 'IB5'], + ['邓白氏AAA+是什么', '邓白氏'], + ['一成系统是什么', '一成系统'], + ['火炉原理是什么', '火炉原理'], + ['D-Drink排毒饮怎么用', 'D-Drink'], + ['怎么加入PM', '加入方式'], + ['孕妇能吃PM产品吗', '特殊人群'], + ['多少钱', '价格'], + ]; + + for (const [query, label] of hotTopics) { + it(`${label}: "${query}" → 应路由到KB（热答案可达）`, () => { + assertKbRoute(query); + }); + } +}); + +// ================================================================ +// 8. 问题维度交叉测试（同一产品不同问法） +// ================================================================ +describe('问题维度交叉 —— 同一产品的不同问法', () => { + + describe('基础三合一 × 多维度', () => { + const dimensions = [ + '基础三合一怎么吃', + '基础三合一功效', + '基础三合一成分', + '基础三合一多少钱', + '基础三合一适合谁', + '为什么要全套搭配三合一', + ]; + for (const q of dimensions) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } + }); + + describe('小红 × 多维度', () => { + const dimensions = [ + '小红怎么吃', + '小红功效是什么', + '小红成分有哪些', + '小红副作用', + '小红多少钱', + '小红适合什么人', + ]; + for (const q of dimensions) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } + }); + + describe('一成系统 × 多维度', () => { + const dimensions = [ + '一成系统是什么', + '一成系统核心竞争力', + '一成系统怎么用', + '一成系统三大平台', + '一成系统AI智能生产力', + '一成系统线上拓客', + '一成系统邀约话术', + '一成系统文化', + ]; + for (const q of dimensions) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } + }); +}); + +// ================================================================ +// 9. 口语/ASR变体测试（语音识别的常见错误变体） +// ================================================================ +describe('口语/ASR变体 —— 语音识别常见变体', () => { + const asrVariants = [ + ['移程系统', '一成系统ASR变体'], + ['PM细胞营养素', 'PM产品'], + ['暖炉原理', '火炉原理ASR变体'], + ['产品有哪些', '产品列表'], + ['你们公司产品', '口语化'], + ['这个东西怎么吃', '口语化追问'], + ['咱们公司介绍一下', '口语化公司'], + ]; + + for (const [query, label] of asrVariants) { + it(`${label}: "${query}" → 应走KB路由`, () => { + assertKbRoute(query); + }); + } +}); + +// ================================================================ +// 10. 边界测试 +// ================================================================ +describe('边界测试', () => { + it('空字符串 → 不走KB', () => { + assertNotKbRoute(''); + }); + + it('纯标点 → 不走KB', () => { + assertNotKbRoute('？？？'); + }); + + it('单字"好" → 不走KB', () => { + assertNotKbRoute('好'); + }); + + it('纯数字 → 不走KB', () => { + assertNotKbRoute('123456'); + }); + + it('超长无意义文本 → 不走KB', () => { + assertNotKbRoute('啊啊啊啊啊啊啊啊啊啊啊啊啊啊啊啊啊啊啊啊'); + }); + + it('纯英文闲聊 → 不走KB', () => { + assertNotKbRoute('hello how are you'); + }); + + it('含KB关键词但实际是闲聊的边界', () => { + // "不对"是质疑词，会被检测为KB follow-up + // 需要context才会真正路由到KB + // 无context时: hasKnowledgeRouteKeyword('不对') → true (因为加了质疑词) + // 这是预期行为：宁可多查一次KB，也不要漏掉用户质疑 + const result = shouldForceKnowledgeRoute('不对'); + assert.equal(typeof result, 'boolean', 'Should return boolean'); + }); +}); + +console.log('\n=== KB场景测试加载完成 ===\n'); diff --git a/test2/server/tests/test_kb_scenarios_extended.js b/test2/server/tests/test_kb_scenarios_extended.js new file mode 100644 index 0000000..63b6b04 --- /dev/null +++ b/test2/server/tests/test_kb_scenarios_extended.js @@ -0,0 +1,491 @@ +/** + * KB场景深度扩展测试 + * 覆盖：更多产品×维度交叉、更多确定性改写、更多追问变体、更多多轮切换、更多热答案、更多边界 + * + * 运行方式: node --test tests/test_kb_scenarios_extended.js + */ +const { describe, it } = require('node:test'); +const assert = require('node:assert/strict'); + +const { shouldForceKnowledgeRoute } = require('../services/realtimeDialogRouting'); +const { hasKnowledgeRouteKeyword } = require('../services/knowledgeKeywords'); +const ToolExecutor = require('../services/toolExecutor'); + +function assertKbRoute(text, ctx, msg) { + assert.equal(shouldForceKnowledgeRoute(text, ctx), true, msg || `"${text}" should route to KB`); +} +function assertNotKbRoute(text, ctx, msg) { + assert.equal(shouldForceKnowledgeRoute(text, ctx), false, msg || `"${text}" should NOT route to KB`); +} +function buildCtx(pairs) { + return pairs.map(([role, content]) => ({ role, content })); +} + +// ================================================================ +// 1. 产品×维度全交叉 —— 每个产品 × 每个问题维度 +// ================================================================ +describe('产品×维度全交叉 —— 大白', () => { + const dims = ['怎么吃', '功效是什么', '成分有哪些', '多少钱', '适合谁', '副作用', '哪里买', '保质期']; + for (const d of dims) { + it(`"大白${d}" → 应走KB`, () => { assertKbRoute(`大白${d}`); }); + } +}); + +describe('产品×维度全交叉 —— 小白', () => { + const dims = ['怎么吃', '功效是什么', '成分有哪些', '多少钱', '适合谁', '副作用', '哪里买', '怎么服用']; + for (const d of dims) { + it(`"小白${d}" → 应走KB`, () => { assertKbRoute(`小白${d}`); }); + } +}); + +describe('产品×维度全交叉 —— CC套装', () => { + const dims = ['怎么用', '功效是什么', '成分有哪些', '多少钱', '适合谁', '副作用', '区别', '包含什么']; + for (const d of dims) { + it(`"CC套装${d}" → 应走KB`, () => { assertKbRoute(`CC套装${d}`); }); + } +}); + +describe('产品×维度全交叉 —— Q10', () => { + const dims = ['怎么吃', '功效是什么', '成分', '多少钱', '适合谁', '副作用', '怎么买', '适合什么人']; + for (const d of dims) { + it(`"Q10${d}" → 应走KB`, () => { assertKbRoute(`Q10${d}`); }); + } +}); + +describe('产品×维度全交叉 —— IB5', () => { + const dims = ['怎么用', '功效是什么', '成分', '多少钱', '适合谁', '副作用', '什么时候用', '适合什么人']; + for (const d of dims) { + it(`"IB5${d}" → 应走KB`, () => { assertKbRoute(`IB5${d}`); }); + } +}); + +describe('产品×维度全交叉 —— D-Drink', () => { + const dims = ['怎么用', '功效是什么', '成分', '多少钱', '适合谁', '副作用', '排毒原理', '喝法']; + for (const d of dims) { + it(`"D-Drink${d}" → 应走KB`, () => { assertKbRoute(`D-Drink${d}`); }); + } +}); + +describe('产品×维度全交叉 —— Hair+', () => { + const dims = ['怎么用', '功效是什么', '成分', '多少钱', '适合谁', '副作用']; + for (const d of dims) { + it(`"Hair+${d}" → 应走KB`, () => { assertKbRoute(`Hair+${d}`); }); + } +}); + +describe('产品×维度全交叉 —— 儿童倍适', () => { + const dims = ['怎么吃', '功效是什么', '成分', '多少钱', '适合几岁', '副作用', '适合什么人', '口味']; + for (const d of dims) { + it(`"儿童倍适${d}" → 应走KB`, () => { assertKbRoute(`儿童倍适${d}`); }); + } +}); + +describe('产品×维度全交叉 —— 关节套装', () => { + const dims = ['怎么用', '功效是什么', '成分', '多少钱', '适合谁', '副作用']; + for (const d of dims) { + it(`"关节套装${d}" → 应走KB`, () => { assertKbRoute(`关节套装${d}`); }); + } +}); + +// ================================================================ +// 2. 确定性改写深度覆盖 —— 更多产品规则 +// ================================================================ +describe('确定性改写深度 —— 更多产品规则', () => { + + describe('一成系统及子话题', () => { + const cases = [ + ['一成系统是什么', '一成系统'], + ['一成系统怎么用', '一成系统'], + ['一成系统三大平台', '一成系统'], + ['一成系统行动圈', '一成系统'], + ['身未动梦已成', '一成系统'], + ['一部手机做天下', '一成系统'], + ['如何发展PM事业', '一成系统'], + ]; + for (const [q, expect] of cases) { + it(`"${q}" → 改写含"${expect}"`, () => { + const r = ToolExecutor.buildDeterministicKnowledgeQuery(q, []); + assert.ok(r && r.includes(expect), `"${q}" → got "${r}"`); + }); + } + }); + + describe('PM公司相关', () => { + const cases = [ + ['德国PM公司介绍', 'PM'], + ['PM公司背景', 'PM'], + ['PM是不是传销', 'PM'], + ['PM公司合法吗', 'PM'], + ]; + for (const [q, expect] of cases) { + it(`"${q}" → 改写含"${expect}"`, () => { + const r = ToolExecutor.buildDeterministicKnowledgeQuery(q, []); + assert.ok(r && r.includes(expect), `"${q}" → got "${r}"`); + }); + } + }); + + describe('NTC/火炉原理/阿育吠陀', () => { + const cases = [ + ['NTC营养保送系统是什么', 'NTC'], + ['NTC核心优势', 'NTC'], + ['火炉原理', '火炉原理'], + ['阿育吠陀是什么', '阿育吠陀'], + ]; + for (const [q, expect] of cases) { + it(`"${q}" → 改写含"${expect}"`, () => { + const r = ToolExecutor.buildDeterministicKnowledgeQuery(q, []); + assert.ok(r && r.includes(expect), `"${q}" → got "${r}"`); + }); + } + }); + + describe('上下文追问改写 —— 各产品', () => { + const ctxCases = [ + [{ role: 'assistant', content: '小红Activize是...' }, '成分是什么', 'Activize'], + [{ role: 'assistant', content: '小白Restorate帮助修复...' }, '适合谁', 'Restorate'], + [{ role: 'assistant', content: 'CC套装含有CC-Cell葡萄籽精华胶囊和乳霜' }, '怎么吃', 'CC'], + [{ role: 'assistant', content: 'Hair+发宝防脱发口服发宝外用发健' }, '功效', 'Hair'], + [{ role: 'assistant', content: '儿童倍适PowerCocktail Junior适合小朋友' }, '怎么用', '儿童倍适'], + [{ role: 'assistant', content: 'D-Drink小绿排毒饮是14天排毒方案' }, '功效是什么', 'D-Drink'], + [{ role: 'assistant', content: 'Apple Antioxy Zellschutz细胞抗氧素是独立小袋包装' }, '他的规格是什么', 'Apple Antioxy'], + [{ role: 'assistant', content: '小白Restorate建议睡前空腹服用' }, '它一天几次', 'Restorate'], + ]; + for (const [ctxMsg, query, expect] of ctxCases) { + it(`上下文"${ctxMsg.content.slice(0, 10)}..."追问"${query}" → 含"${expect}"`, () => { + const r = ToolExecutor.buildDeterministicKnowledgeQuery(query, [ctxMsg]); + assert.ok(r && r.includes(expect), `Got "${r}"`); + }); + } + }); + + describe('无匹配场景', () => { + const noMatch = ['你好', '天气怎么样', '讲个故事', '几点了', '我要听音乐']; + for (const q of noMatch) { + it(`"${q}" → 无改写`, () => { + const r = ToolExecutor.buildDeterministicKnowledgeQuery(q, []); + assert.equal(r, '', `"${q}" should not rewrite, got "${r}"`); + }); + } + }); +}); + +// ================================================================ +// 3. 追问变体全覆盖 —— 各种追问句式 × 不同产品上下文 +// ================================================================ +describe('追问变体全覆盖 —— 不同追问句式', () => { + const followUpPatterns = [ + '怎么吃', '怎么用', '功效是什么', '成分是什么', '多少钱', + '适合谁', '哪里买', '什么意思', '有什么好处', '怎么服用', + '详细说说', '介绍一下', '继续说', '展开说说', '配方', + '原理是什么', '适合什么人', '怎么买', '具体内容', + ]; + + const ctxProducts = [ + buildCtx([['user', '大白产品'], ['assistant', '大白Basics...']]), + buildCtx([['user', '小红功效'], ['assistant', '小红Activize...']]), + buildCtx([['user', '一成系统'], ['assistant', '一成系统是...']]), + ]; + + for (let pi = 0; pi < ctxProducts.length; pi++) { + for (const fup of followUpPatterns) { + it(`ctx${pi + 1} + "${fup}" → 应走KB`, () => { + assertKbRoute(fup, ctxProducts[pi]); + }); + } + } +}); + +// ================================================================ +// 4. 多KB切换 —— 更多组合 +// ================================================================ +describe('多KB切换 —— 更多组合模式', () => { + + it('产品→FAQ→科学: 小红→副作用→NTC原理', () => { + assertKbRoute('小红有什么副作用'); + assertKbRoute('PM产品有副作用吗'); + assertKbRoute('NTC营养保送系统原理'); + }); + + it('事业→公司→产品→FAQ: 招商→PM背景→大白→见效时间', () => { + assertKbRoute('招商代理政策'); + assertKbRoute('德国PM公司背景'); + assertKbRoute('大白怎么吃'); + assertKbRoute('多久见效'); + }); + + it('科学→产品→产品→产品: NTC→小红→小白→CC', () => { + assertKbRoute('NTC是什么'); + assertKbRoute('小红功效'); + assertKbRoute('小白成分'); + assertKbRoute('CC套装怎么用'); + }); + + it('FAQ→FAQ→FAQ: 传销→副作用→见效→全套搭配', () => { + assertKbRoute('PM是传销吗'); + assertKbRoute('PM产品有副作用吗'); + assertKbRoute('多久能见效'); + assertKbRoute('为什么要全套搭配'); + }); + + it('系统→事业→认证: 一成系统→做PM→邓白氏', () => { + assertKbRoute('一成系统介绍'); + assertKbRoute('怎么做PM事业'); + assertKbRoute('邓白氏AAA+认证'); + }); + + it('产品→追问→切换→追问: 大白→怎么吃→小红→功效', () => { + assertKbRoute('大白是什么'); + const ctx1 = buildCtx([['user', '大白'], ['assistant', '大白Basics...']]); + assertKbRoute('怎么吃', ctx1); + assertKbRoute('小红产品介绍'); + const ctx2 = buildCtx([['user', '小红'], ['assistant', '小红Activize...']]); + assertKbRoute('功效是什么', ctx2); + }); +}); + +// ================================================================ +// 5. 质疑+追问混合 —— 更多组合场景 +// ================================================================ +describe('质疑+追问混合 —— 更多组合', () => { + + it('质疑后继续追问详情', () => { + assertKbRoute('大白功效'); + const ctx = buildCtx([['user', '大白功效'], ['assistant', '大白Basics帮助...']]); + assertKbRoute('你搞错了吧', ctx); + assertKbRoute('那正确的功效到底是什么', ctx); + }); + + it('追问后发现错误再质疑', () => { + assertKbRoute('CC套装包含什么'); + const ctx = buildCtx([['user', 'CC套装'], ['assistant', 'CC套装含有...']]); + assertKbRoute('具体成分是什么', ctx); + assertKbRoute('说的有问题，我记得不是这个成分', ctx); + }); + + it('质疑→纠正→再追问另一个维度', () => { + const ctx = buildCtx([['user', '小红怎么吃'], ['assistant', '小红冲服...']]); + assertKbRoute('不对，不是冲着喝的', ctx); + assertKbRoute('小红到底是什么剂型', ctx); + assertKbRoute('小红适合什么人群', ctx); + }); + + it('连续切换产品并质疑', () => { + assertKbRoute('大白功效'); + assertKbRoute('你说的不对'); + assertKbRoute('小红功效'); + assertKbRoute('也不对吧'); + assertKbRoute('小白功效'); + assertKbRoute('说的不准确'); + }); + + it('先闲聊再切到KB质疑', () => { + assertNotKbRoute('今天心情不错'); + assertKbRoute('对了基础三合一怎么吃'); + const ctx = buildCtx([['user', '基础三合一'], ['assistant', '大白小红小白...']]); + assertKbRoute('不是这样吧', ctx); + }); +}); + +// ================================================================ +// 6. 热答案可达性 —— 更多热门问题变体 +// ================================================================ +describe('热答案可达性 —— 问题变体', () => { + const hotVariants = [ + // 基础三合一吃法的各种问法 + '基础三合一怎么吃', '基础三合一的吃法', '基础套装怎么服用', + '大白小红小白怎么吃', + // 传销/合法性的各种问法 + 'PM是不是传销', 'PM合法吗', 'PM是传销吗', 'PM正规吗', + // NTC + 'NTC是什么', 'NTC核心优势是什么', 'NTC营养保送系统', + // 见效时间 + '多久见效', '多久能见效', '吃多久有效果', + // 好转反应 + '好转反应', '好转反应是什么', '吃了不舒服正常吗', + // 公司 + '德国PM公司介绍', 'PM公司背景', 'PM公司怎么样', + // 价格 + '多少钱', '产品价格', '产品贵不贵', + ]; + + for (const q of hotVariants) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } +}); + +// ================================================================ +// 7. 口语化查询 —— 更多自然说法 +// ================================================================ +describe('口语化查询 —— 真实用户的自然说法', () => { + const colloquial = [ + '你们公司是做什么的', + '你们那个产品怎么吃', + '咱们这个东西多少钱', + '那个什么三合一是什么', + '帮我介绍一下你们产品', + '我想了解PM产品', + '说说你们公司', + '讲讲那个什么系统', + '查查那个产品', + '帮我查一下基础三合一', + '帮我问一下价格', + '你们产品正规吗', + '你们那个东西靠谱吗', + '说说那个什么功效', + '我想知道怎么加入', + '你们卖的是什么东西', + '健康产品有哪些', + '帮我看看成分', + '你们的东西有什么用', + '咱吃这个有好处吗', + ]; + + for (const q of colloquial) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } +}); + +// ================================================================ +// 8. 负面边界 —— 更多不应走KB的场景 +// ================================================================ +describe('负面边界 —— 更多不应走KB的场景', () => { + const negatives = [ + '', + ' ', + '？', + '！！！', + '。。。', + '好', + '嗯', + '哦', + '行', + '啊', + '对', + '是的', + '好的好的', + '知道了知道了', + '哈哈哈哈', + '嘻嘻', + '666', + '999', + '111', + '早', + '晚安', + 'ok', + 'OK', + 'hello', + 'hi', + 'bye', + 'good', + '今天天气真好', + '我要睡觉了', + '你是AI吗', + '你能做什么', + ]; + + for (const text of negatives) { + it(`"${text}" → 不应走KB`, () => { assertNotKbRoute(text); }); + } +}); + +// ================================================================ +// 9. 关键词嵌入长句 —— 确保关键词在句中也能命中 +// ================================================================ +describe('关键词嵌入长句 —— 子串匹配验证', () => { + const embedded = [ + ['我想了解一下基础三合一的功效', '基础三合一嵌入'], + ['请问德国PM公司在哪个城市', '公司嵌入'], + ['你能帮我查查一成系统怎么用吗', '一成系统嵌入'], + ['我朋友推荐我吃小红你能介绍下吗', '小红嵌入'], + ['听说有个叫NTC的营养保送系统', 'NTC嵌入'], + ['好转反应的话应该怎么处理', '好转反应嵌入'], + ['孕妇是不是不能吃PM的产品', '孕妇嵌入'], + ['为什么说要全套搭配使用呢', '全套搭配嵌入'], + ['听人家说邓白氏AAA+评级很厉害', '邓白氏嵌入'], + ['想问一下儿童倍适几岁能吃', '儿童倍适嵌入'], + ['我对火炉原理很感兴趣', '火炉原理嵌入'], + ['请介绍一下CC套装的功效', 'CC套装嵌入'], + ['Q10辅酵素是做什么用的', 'Q10嵌入'], + ['Hair+发宝真的能防脱吗', 'Hair+嵌入'], + ['想知道D-Drink排毒的原理', 'D-Drink嵌入'], + ]; + + for (const [text, label] of embedded) { + it(`${label}: "${text}" → 应走KB`, () => { assertKbRoute(text); }); + } +}); + +// ================================================================ +// 10. 同义问法覆盖 —— 相同意图不同表达 +// ================================================================ +describe('同义问法 —— 相同意图不同表达', () => { + + describe('询问功效的N种方式', () => { + const efficacyCases = [ + '大白有什么功效', '大白的作用是什么', '大白有什么用', + '大白好在哪', '吃大白有什么好处', + ]; + for (const q of efficacyCases) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } + }); + + describe('询问用法的N种方式', () => { + const usageCases = [ + '小红怎么吃', '小红怎么服用', '小红怎么用', + '小红的吃法', '小红用法', + ]; + for (const q of usageCases) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } + }); + + describe('询问价格的N种方式', () => { + const priceCases = [ + '基础三合一多少钱', '基础三合一价格', '基础三合一贵不贵', + '产品多少钱', '产品价格表', + ]; + for (const q of priceCases) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } + }); + + describe('询问合法性的N种方式', () => { + const legalCases = [ + 'PM是不是传销', 'PM正规吗', 'PM合法吗', + 'PM是传销吗', 'PM是直销还是传销', 'PM靠谱吗', + ]; + for (const q of legalCases) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } + }); + + describe('询问公司的N种方式', () => { + const companyCases = [ + 'PM公司介绍', '德国PM公司怎么样', + 'PM公司成立多久了', 'PM公司实力如何', + ]; + for (const q of companyCases) { + it(`"${q}" → 应走KB`, () => { assertKbRoute(q); }); + } + }); +}); + +// ================================================================ +// 11. 特殊人群×产品交叉 +// ================================================================ +describe('特殊人群×产品 —— 能不能吃的场景', () => { + const groups = ['孕妇', '小孩', '老人', '糖尿病人', '高血压']; + const products = ['大白', '小红', '基础三合一', 'PM产品']; + + for (const g of groups) { + for (const p of products) { + it(`"${g}能吃${p}吗" → 应走KB`, () => { + assertKbRoute(`${g}能吃${p}吗`); + }); + } + } +}); + +console.log('\n=== KB场景扩展测试加载完成 ===\n'); diff --git a/test2/server/tests/test_real_viking_kb.js b/test2/server/tests/test_real_viking_kb.js new file mode 100644 index 0000000..a991dd1 --- /dev/null +++ b/test2/server/tests/test_real_viking_kb.js @@ -0,0 +1,112 @@ +const path = require('path'); +const { performance } = require('perf_hooks'); + +require('dotenv').config({ path: path.join(__dirname, '../.env') }); + +const ToolExecutor = require('../services/toolExecutor'); + +const realTestQueries = [ + { name: 'Product - Q10', query: 'Q10辅酵素氧修护有什么作用' }, + { name: 'Product - IB5', query: 'IB5口腔免疫喷雾怎么使用' }, + { name: 'Product - CC胶囊', query: 'CC胶囊适合什么人使用' }, + { name: 'Company - 邓白氏', query: '德国PM的邓白氏认证是多少分' }, + { name: 'Technology - 火炉原理', query: '请详细解释一下火炉原理' }, + { name: 'Training - 新人起步', query: '培训新人起步三关是什么' }, + { name: 'Product - 关节套装', query: '关节套装有什么功效' }, + { name: 'Company - 培安烟台', query: '培安烟台是什么' }, + { name: 'Product - 儿童倍适', query: '儿童倍适有什么成分' }, + { name: 'Science - 阿育吠陀', query: '阿育吠陀医学原理是什么' } +]; + +async function runRealKBTest() { + console.log('='.repeat(80)); + console.log('VIKING REAL KNOWLEDGE BASE PERFORMANCE TEST'); + console.log('='.repeat(80)); + console.log(''); + + console.log('Warming up...'); + for (let i = 0; i < 2; i++) { + for (const q of realTestQueries) { + try { + await ToolExecutor.searchKnowledge({ query: q.query, response_mode: 'answer' }, []); + } catch (e) {} + } + } + console.log('Warmup complete!\n'); + + console.log('Running latency tests (5 iterations each)...\n'); + + const allResults = []; + + for (const { name, query } of realTestQueries) { + console.log(`Testing: ${name}`); + const latencies = []; + const hits = []; + + for (let i = 0; i < 5; i++) { + const start = performance.now(); + let result; + try { + result = await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + const latency = performance.now() - start; + latencies.push(latency); + hits.push(!!result.hit); + console.log(` Iteration ${i + 1}: ${latency.toFixed(2)}ms, hit=${result.hit}, source=${result.source}`); + } catch (e) { + console.log(` Iteration ${i + 1} error: ${e.message}`); + } + } + + const avgLatency = latencies.length ? latencies.reduce((a, b) => a + b, 0) / latencies.length : 0; + const p50 = percentile(latencies, 50); + const p95 = percentile(latencies, 95); + const hitRate = hits.length ? hits.filter(h => h).length / hits.length : 0; + + allResults.push({ + name, + query, + avgLatency, + p50, + p95, + hitRate, + latencies + }); + + console.log(` → Avg: ${avgLatency.toFixed(2)}ms, P95: ${p95.toFixed(2)}ms, Hit Rate: ${(hitRate * 100).toFixed(1)}%\n`); + } + + console.log('='.repeat(80)); + console.log('SUMMARY'); + console.log('='.repeat(80)); + + const totalAvg = allResults.reduce((a, b) => a + b.avgLatency, 0) / allResults.length; + const totalP95 = percentile(allResults.flatMap(r => r.latencies), 95); + const totalHitRate = allResults.reduce((a, b) => a + b.hitRate, 0) / allResults.length; + + console.log(`\nOverall Average Latency: ${totalAvg.toFixed(2)}ms`); + console.log(`Overall P95 Latency: ${totalP95.toFixed(2)}ms`); + console.log(`Overall Hit Rate: ${(totalHitRate * 100).toFixed(1)}%`); + + console.log('\nTop 3 fastest queries:'); + allResults.sort((a, b) => a.avgLatency - b.avgLatency).slice(0, 3).forEach((r, i) => { + console.log(` ${i + 1}. ${r.name}: ${r.avgLatency.toFixed(2)}ms`); + }); + + console.log('\nTop 3 slowest queries:'); + allResults.sort((a, b) => b.avgLatency - a.avgLatency).slice(0, 3).forEach((r, i) => { + console.log(` ${i + 1}. ${r.name}: ${r.avgLatency.toFixed(2)}ms`); + }); + + console.log('\n' + '='.repeat(80)); + + return allResults; +} + +function percentile(arr, p) { + if (arr.length === 0) return 0; + const sorted = [...arr].sort((a, b) => a - b); + const index = Math.ceil((p / 100) * sorted.length) - 1; + return sorted[Math.max(0, index)]; +} + +runRealKBTest().catch(console.error); diff --git a/test2/server/tests/test_results/viking_performance_1773992461552.json b/test2/server/tests/test_results/viking_performance_1773992461552.json new file mode 100644 index 0000000..c978770 --- /dev/null +++ b/test2/server/tests/test_results/viking_performance_1773992461552.json @@ -0,0 +1,498 @@ +{ + "generatedAt": "2026-03-20T07:41:01.552Z", + "mockMode": true, + "summary": { + "latency": { + "Product Query - Xiaohong": { + "avg": "221.59", + "p95": "313.74", + "hitRate": "100.0%" + }, + "Product Query - Dabai": { + "avg": "173.13", + "p95": "291.55", + "hitRate": "100.0%" + }, + "Company Info": { + "avg": "176.99", + "p95": "244.86", + "hitRate": "100.0%" + }, + "NTC Technology": { + "avg": "197.35", + "p95": "295.00", + "hitRate": "100.0%" + }, + "Hot Answer": { + "avg": "209.49", + "p95": "279.63", + "hitRate": "100.0%" + }, + "No Hit Query": { + "avg": "204.48", + "p95": "296.13", + "hitRate": "0.0%" + } + }, + "cache": [ + { + "name": "Product Query - Xiaohong", + "speedup": "0.50x" + }, + { + "name": "Product Query - Dabai", + "speedup": "0.79x" + }, + { + "name": "Company Info", + "speedup": "0.86x" + } + ], + "concurrency": { + "1": { + "throughput": "8.00 req/s", + "successRate": "100.0%" + }, + "3": { + "throughput": "11.41 req/s", + "successRate": "100.0%" + }, + "5": { + "throughput": "16.96 req/s", + "successRate": "100.0%" + } + } + }, + "tests": [ + { + "type": "latency", + "timestamp": "2026-03-20T07:40:56.968Z", + "iterations": 10, + "mockMode": true, + "results": { + "Product Query - Xiaohong": { + "query": "小红产品有什么功效", + "latencies": [ + 215.31479999999965, + 248.87959999999975, + 313.73730000000023, + 216.14289999999983, + 107.47650000000021, + 122.7501999999995, + 185.82290000000012, + 216.10750000000007, + 294.0571, + 295.65119999999933 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 221.59399999999988, + "minLatency": 107.47650000000021, + "maxLatency": 313.73730000000023, + "p50Latency": 216.10750000000007, + "p95Latency": 313.73730000000023, + "p99Latency": 313.73730000000023, + "hitRate": 1 + }, + "Product Query - Dabai": { + "query": "大白产品怎么吃", + "latencies": [ + 108.59429999999975, + 109.21360000000004, + 189.07920000000013, + 141.89410000000044, + 264.7583999999997, + 123.83699999999953, + 291.5468000000001, + 188.14480000000003, + 203.59680000000026, + 110.61819999999989 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 173.12831999999997, + "minLatency": 108.59429999999975, + "maxLatency": 291.5468000000001, + "p50Latency": 141.89410000000044, + "p95Latency": 291.5468000000001, + "p99Latency": 291.5468000000001, + "hitRate": 1 + }, + "Company Info": { + "query": "德国PM公司介绍", + "latencies": [ + 122.88140000000021, + 140.60269999999946, + 170.5096999999996, + 140.1189000000004, + 187.21970000000056, + 218.0506000000005, + 220.02800000000025, + 244.85799999999927, + 124.36050000000068, + 201.2597999999998 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 176.98893000000007, + "minLatency": 122.88140000000021, + "maxLatency": 244.85799999999927, + "p50Latency": 170.5096999999996, + "p95Latency": 244.85799999999927, + "p99Latency": 244.85799999999927, + "hitRate": 1 + }, + "NTC Technology": { + "query": "NTC营养保送系统原理", + "latencies": [ + 201.2448000000004, + 294.9982, + 123.97750000000087, + 217.60050000000047, + 250.01519999999982, + 154.38770000000113, + 123.49189999999908, + 294.09230000000025, + 140.3801999999996, + 173.35200000000077 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 197.35403000000025, + "minLatency": 123.49189999999908, + "maxLatency": 294.9982, + "p50Latency": 173.35200000000077, + "p95Latency": 294.9982, + "p99Latency": 294.9982, + "hitRate": 1 + }, + "Hot Answer": { + "query": "基础三合一怎么吃", + "latencies": [ + 185.3747000000003, + 279.6288999999997, + 139.79800000000068, + 278.0802999999996, + 124.75540000000001, + 278.8053999999993, + 139.687100000001, + 234.39990000000034, + 202.0474000000013, + 232.36249999999927 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 209.49396000000016, + "minLatency": 124.75540000000001, + "maxLatency": 279.6288999999997, + "p50Latency": 202.0474000000013, + "p95Latency": 279.6288999999997, + "p99Latency": 279.6288999999997, + "hitRate": 1 + }, + "No Hit Query": { + "query": "今天天气怎么样", + "latencies": [ + 216.4937000000009, + 202.85559999999896, + 122.8827999999994, + 248.70260000000053, + 186.36679999999978, + 140.94870000000083, + 296.1263999999992, + 246.18019999999888, + 230.27289999999994, + 153.9431000000004 + ], + "hits": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "avgLatency": 204.4772799999999, + "minLatency": 122.8827999999994, + "maxLatency": 296.1263999999992, + "p50Latency": 202.85559999999896, + "p95Latency": 296.1263999999992, + "p99Latency": 296.1263999999992, + "hitRate": 0 + } + } + }, + { + "type": "cache", + "timestamp": "2026-03-20T07:41:00.864Z", + "cacheHitsIterations": 5, + "mockMode": true, + "results": [ + { + "name": "Product Query - Xiaohong", + "query": "小红产品有什么功效", + "firstHitLatency": 126.06420000000071, + "cacheHitLatencies": [ + 248.09699999999975, + 292.3535000000011, + 279.5563999999995, + 278.85109999999986, + 168.8062000000009 + ], + "avgCacheLatency": 253.53284000000022, + "speedup": 0.49723026019035876, + "firstHit": { + "query": "小红产品有什么功效", + "results": [ + { + "title": "模拟知识库结果", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。" + } + ], + "hit": true, + "source": "mock_knowledge" + } + }, + { + "name": "Product Query - Dabai", + "query": "大白产品怎么吃", + "firstHitLatency": 171.05249999999796, + "cacheHitLatencies": [ + 282.6729000000014, + 219.4409999999989, + 218.62560000000303, + 139.14129999999932, + 216.64949999999953 + ], + "avgCacheLatency": 215.30606000000043, + "speedup": 0.7944620787728762, + "firstHit": { + "query": "大白产品怎么吃", + "results": [ + { + "title": "模拟知识库结果", + "content": "德国PM大白Basics是基础营养素，早上空腹服用，1平勺兑200-300ml温水。" + } + ], + "hit": true, + "source": "mock_knowledge" + } + }, + { + "name": "Company Info", + "query": "德国PM公司介绍", + "firstHitLatency": 183.53129999999874, + "cacheHitLatencies": [ + 292.8653000000013, + 107.31589999999778, + 233.82210000000123, + 262.02170000000115, + 170.1095000000023 + ], + "avgCacheLatency": 213.22690000000074, + "speedup": 0.8607323935206961, + "firstHit": { + "query": "德国PM公司介绍", + "results": [ + { + "title": "模拟知识库结果", + "content": "德国PM-International是1993年创立的国际营养品直销企业。核心品牌FitLine专注细胞营养，拥有独家NTC营养保送系统技术。" + } + ], + "hit": true, + "source": "mock_knowledge" + } + } + ] + }, + { + "type": "concurrency", + "timestamp": "2026-03-20T07:41:01.549Z", + "concurrencyLevels": [ + 1, + 3, + 5 + ], + "mockMode": true, + "results": { + "1": { + "concurrency": 1, + "totalTime": 125.04439999999886, + "throughput": 7.997159408978004, + "successRate": 1, + "results": [ + { + "query": "小红产品有什么功效", + "results": [ + { + "title": "模拟知识库结果", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。" + } + ], + "hit": true, + "source": "mock_knowledge" + } + ] + }, + "3": { + "concurrency": 3, + "totalTime": 262.9516000000003, + "throughput": 11.408943699144618, + "successRate": 1, + "results": [ + { + "query": "小红产品有什么功效", + "results": [ + { + "title": "模拟知识库结果", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。" + } + ], + "hit": true, + "source": "mock_knowledge" + }, + { + "query": "大白产品怎么吃", + "results": [ + { + "title": "模拟知识库结果", + "content": "德国PM大白Basics是基础营养素，早上空腹服用，1平勺兑200-300ml温水。" + } + ], + "hit": true, + "source": "mock_knowledge" + }, + { + "query": "德国PM公司介绍", + "results": [ + { + "title": "模拟知识库结果", + "content": "德国PM-International是1993年创立的国际营养品直销企业。核心品牌FitLine专注细胞营养，拥有独家NTC营养保送系统技术。" + } + ], + "hit": true, + "source": "mock_knowledge" + } + ] + }, + "5": { + "concurrency": 5, + "totalTime": 294.8380999999972, + "throughput": 16.95845957493298, + "successRate": 1, + "results": [ + { + "query": "小红产品有什么功效", + "results": [ + { + "title": "模拟知识库结果", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。" + } + ], + "hit": true, + "source": "mock_knowledge" + }, + { + "query": "大白产品怎么吃", + "results": [ + { + "title": "模拟知识库结果", + "content": "德国PM大白Basics是基础营养素，早上空腹服用，1平勺兑200-300ml温水。" + } + ], + "hit": true, + "source": "mock_knowledge" + }, + { + "query": "德国PM公司介绍", + "results": [ + { + "title": "模拟知识库结果", + "content": "德国PM-International是1993年创立的国际营养品直销企业。核心品牌FitLine专注细胞营养，拥有独家NTC营养保送系统技术。" + } + ], + "hit": true, + "source": "mock_knowledge" + }, + { + "query": "小红产品有什么功效", + "results": [ + { + "title": "模拟知识库结果", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。" + } + ], + "hit": true, + "source": "mock_knowledge" + }, + { + "query": "大白产品怎么吃", + "results": [ + { + "title": "模拟知识库结果", + "content": "德国PM大白Basics是基础营养素，早上空腹服用，1平勺兑200-300ml温水。" + } + ], + "hit": true, + "source": "mock_knowledge" + } + ] + } + } + } + ] +} \ No newline at end of file diff --git a/test2/server/tests/test_results/viking_performance_1773992947093.json b/test2/server/tests/test_results/viking_performance_1773992947093.json new file mode 100644 index 0000000..81da846 --- /dev/null +++ b/test2/server/tests/test_results/viking_performance_1773992947093.json @@ -0,0 +1,570 @@ +{ + "generatedAt": "2026-03-20T07:49:07.093Z", + "mockMode": false, + "summary": { + "latency": { + "Product Query - Xiaohong": { + "avg": "0.15", + "p95": "0.22", + "hitRate": "100.0%" + }, + "Product Query - Dabai": { + "avg": "0.15", + "p95": "0.17", + "hitRate": "100.0%" + }, + "Company Info": { + "avg": "0.13", + "p95": "0.15", + "hitRate": "100.0%" + }, + "NTC Technology": { + "avg": "0.13", + "p95": "0.19", + "hitRate": "100.0%" + }, + "Hot Answer": { + "avg": "0.13", + "p95": "0.14", + "hitRate": "100.0%" + }, + "No Hit Query": { + "avg": "0.66", + "p95": "0.87", + "hitRate": "100.0%" + } + }, + "cache": [ + { + "name": "Product Query - Xiaohong", + "speedup": "1.11x" + }, + { + "name": "Product Query - Dabai", + "speedup": "0.99x" + }, + { + "name": "Company Info", + "speedup": "1.25x" + } + ], + "concurrency": { + "1": { + "throughput": "6016.85 req/s", + "successRate": "100.0%" + }, + "3": { + "throughput": "6426.74 req/s", + "successRate": "100.0%" + }, + "5": { + "throughput": "6439.15 req/s", + "successRate": "100.0%" + } + } + }, + "tests": [ + { + "type": "latency", + "timestamp": "2026-03-20T07:49:07.084Z", + "iterations": 10, + "mockMode": false, + "results": { + "Product Query - Xiaohong": { + "query": "小红产品有什么功效", + "latencies": [ + 0.2159999999998945, + 0.17509999999992942, + 0.1505000000001928, + 0.14269999999987704, + 0.15470000000004802, + 0.1368999999999687, + 0.12730000000010477, + 0.13599999999996726, + 0.13009999999985666, + 0.15689999999995052 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 0.15261999999997897, + "minLatency": 0.12730000000010477, + "maxLatency": 0.2159999999998945, + "p50Latency": 0.14269999999987704, + "p95Latency": 0.2159999999998945, + "p99Latency": 0.2159999999998945, + "hitRate": 1 + }, + "Product Query - Dabai": { + "query": "大白产品怎么吃", + "latencies": [ + 0.15099999999983993, + 0.14449999999987995, + 0.1378999999997177, + 0.13919999999961874, + 0.171100000000024, + 0.14480000000003201, + 0.14390000000003056, + 0.14840000000003783, + 0.14489999999977954, + 0.13879999999971915 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 0.14644999999986794, + "minLatency": 0.1378999999997177, + "maxLatency": 0.171100000000024, + "p50Latency": 0.14449999999987995, + "p95Latency": 0.171100000000024, + "p99Latency": 0.171100000000024, + "hitRate": 1 + }, + "Company Info": { + "query": "德国PM公司介绍", + "latencies": [ + 0.13020000000005894, + 0.12360000000035143, + 0.14609999999993306, + 0.13189999999985957, + 0.13970000000017535, + 0.12660000000005311, + 0.12400000000025102, + 0.11830000000009022, + 0.11290000000008149, + 0.13209999999980937 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 0.12854000000006635, + "minLatency": 0.11290000000008149, + "maxLatency": 0.14609999999993306, + "p50Latency": 0.12660000000005311, + "p95Latency": 0.14609999999993306, + "p99Latency": 0.14609999999993306, + "hitRate": 1 + }, + "NTC Technology": { + "query": "NTC营养保送系统原理", + "latencies": [ + 0.12449999999989814, + 0.12459999999964566, + 0.13049999999975626, + 0.10840000000007421, + 0.19180000000005748, + 0.11740000000008877, + 0.12619999999969878, + 0.12239999999974316, + 0.11910000000034415, + 0.12409999999999854 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 0.12889999999993051, + "minLatency": 0.10840000000007421, + "maxLatency": 0.19180000000005748, + "p50Latency": 0.12409999999999854, + "p95Latency": 0.19180000000005748, + "p99Latency": 0.19180000000005748, + "hitRate": 1 + }, + "Hot Answer": { + "query": "基础三合一怎么吃", + "latencies": [ + 0.13679999999976644, + 0.13400000000001455, + 0.13280000000031578, + 0.13779999999997017, + 0.12579999999979918, + 0.13449999999966167, + 0.14399999999977808, + 0.13839999999981956, + 0.125, + 0.1230000000000473 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 0.13320999999991728, + "minLatency": 0.1230000000000473, + "maxLatency": 0.14399999999977808, + "p50Latency": 0.13400000000001455, + "p95Latency": 0.14399999999977808, + "p99Latency": 0.14399999999977808, + "hitRate": 1 + }, + "No Hit Query": { + "query": "今天天气怎么样", + "latencies": [ + 0.7092999999999847, + 0.647899999999936, + 0.869300000000294, + 0.8305000000000291, + 0.7291000000000167, + 0.5709999999999127, + 0.539299999999912, + 0.5781999999999243, + 0.6104999999997744, + 0.5345000000002074 + ], + "hits": [ + true, + true, + true, + true, + true, + true, + true, + true, + true, + true + ], + "avgLatency": 0.6619599999999991, + "minLatency": 0.5345000000002074, + "maxLatency": 0.869300000000294, + "p50Latency": 0.6104999999997744, + "p95Latency": 0.869300000000294, + "p99Latency": 0.869300000000294, + "hitRate": 1 + } + } + }, + { + "type": "cache", + "timestamp": "2026-03-20T07:49:07.088Z", + "cacheHitsIterations": 5, + "mockMode": false, + "results": [ + { + "name": "Product Query - Xiaohong", + "query": "小红产品有什么功效", + "firstHitLatency": 0.1477999999997337, + "cacheHitLatencies": [ + 0.14179999999987558, + 0.1406000000001768, + 0.13950000000022555, + 0.13410000000021682, + 0.10729999999966822 + ], + "avgCacheLatency": 0.13266000000003259, + "speedup": 1.114126338004654, + "firstHit": { + "query": "小红产品有什么功效", + "original_query": "小红产品有什么功效", + "rewritten_query": "小红产品有什么功效", + "results": [ + { + "title": "高频问题快速回答", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分，通过NTC营养保送系统直达细胞。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。很多人喝完当天就能感受到精力明显提升。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + } + }, + { + "name": "Product Query - Dabai", + "query": "大白产品怎么吃", + "firstHitLatency": 0.1453999999998814, + "cacheHitLatencies": [ + 0.13940000000002328, + 0.13879999999971915, + 0.14170000000012806, + 0.14660000000003492, + 0.16849999999976717 + ], + "avgCacheLatency": 0.14699999999993452, + "speedup": 0.9891156462581372, + "firstHit": { + "query": "大白产品怎么吃", + "original_query": "大白产品怎么吃", + "rewritten_query": "大白产品怎么吃", + "results": [ + { + "title": "高频问题快速回答", + "content": "德国PM大白Basics是基础营养素，为身体提供全面的维生素和矿物质基础。它的核心作用：1.补充每日所需的基础营养。2.为细胞提供全面的营养支持。3.搭配小红和小白形成完整的NTC营养循环。建议早上空腹服用，1平勺兑200-300ml温水。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + } + }, + { + "name": "Company Info", + "query": "德国PM公司介绍", + "firstHitLatency": 0.17620000000033542, + "cacheHitLatencies": [ + 0.1550999999999476, + 0.11990000000014334, + 0.15079999999989013, + 0.14349999999967622, + 0.134900000000016 + ], + "avgCacheLatency": 0.14083999999993466, + "speedup": 1.2510650383443422, + "firstHit": { + "query": "德国PM公司介绍", + "original_query": "德国PM公司介绍", + "rewritten_query": "德国PM公司介绍", + "results": [ + { + "title": "高频问题快速回答", + "content": "德国PM-International是1993年创立的国际营养品直销企业。核心品牌FitLine专注细胞营养，拥有独家NTC营养保送系统技术。公司获邓白氏AAA+最高信用认证（99分），业务覆盖全球100多个国家。总部位于德国，在日本、美国、加拿大、香港等地设有分公司。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + } + } + ] + }, + { + "type": "concurrency", + "timestamp": "2026-03-20T07:49:07.090Z", + "concurrencyLevels": [ + 1, + 3, + 5 + ], + "mockMode": false, + "results": { + "1": { + "concurrency": 1, + "totalTime": 0.1661999999996624, + "throughput": 6016.847172094051, + "successRate": 1, + "results": [ + { + "query": "小红产品有什么功效", + "original_query": "小红产品有什么功效", + "rewritten_query": "小红产品有什么功效", + "results": [ + { + "title": "高频问题快速回答", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分，通过NTC营养保送系统直达细胞。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。很多人喝完当天就能感受到精力明显提升。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + } + ] + }, + "3": { + "concurrency": 3, + "totalTime": 0.4667999999996937, + "throughput": 6426.735218513215, + "successRate": 1, + "results": [ + { + "query": "小红产品有什么功效", + "original_query": "小红产品有什么功效", + "rewritten_query": "小红产品有什么功效", + "results": [ + { + "title": "高频问题快速回答", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分，通过NTC营养保送系统直达细胞。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。很多人喝完当天就能感受到精力明显提升。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + }, + { + "query": "大白产品怎么吃", + "original_query": "大白产品怎么吃", + "rewritten_query": "大白产品怎么吃", + "results": [ + { + "title": "高频问题快速回答", + "content": "德国PM大白Basics是基础营养素，为身体提供全面的维生素和矿物质基础。它的核心作用：1.补充每日所需的基础营养。2.为细胞提供全面的营养支持。3.搭配小红和小白形成完整的NTC营养循环。建议早上空腹服用，1平勺兑200-300ml温水。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + }, + { + "query": "德国PM公司介绍", + "original_query": "德国PM公司介绍", + "rewritten_query": "德国PM公司介绍", + "results": [ + { + "title": "高频问题快速回答", + "content": "德国PM-International是1993年创立的国际营养品直销企业。核心品牌FitLine专注细胞营养，拥有独家NTC营养保送系统技术。公司获邓白氏AAA+最高信用认证（99分），业务覆盖全球100多个国家。总部位于德国，在日本、美国、加拿大、香港等地设有分公司。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + } + ] + }, + "5": { + "concurrency": 5, + "totalTime": 0.7764999999999418, + "throughput": 6439.150032196232, + "successRate": 1, + "results": [ + { + "query": "小红产品有什么功效", + "original_query": "小红产品有什么功效", + "rewritten_query": "小红产品有什么功效", + "results": [ + { + "title": "高频问题快速回答", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分，通过NTC营养保送系统直达细胞。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。很多人喝完当天就能感受到精力明显提升。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 1, + "hot_answer": true + }, + { + "query": "大白产品怎么吃", + "original_query": "大白产品怎么吃", + "rewritten_query": "大白产品怎么吃", + "results": [ + { + "title": "高频问题快速回答", + "content": "德国PM大白Basics是基础营养素，为身体提供全面的维生素和矿物质基础。它的核心作用：1.补充每日所需的基础营养。2.为细胞提供全面的营养支持。3.搭配小红和小白形成完整的NTC营养循环。建议早上空腹服用，1平勺兑200-300ml温水。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + }, + { + "query": "德国PM公司介绍", + "original_query": "德国PM公司介绍", + "rewritten_query": "德国PM公司介绍", + "results": [ + { + "title": "高频问题快速回答", + "content": "德国PM-International是1993年创立的国际营养品直销企业。核心品牌FitLine专注细胞营养，拥有独家NTC营养保送系统技术。公司获邓白氏AAA+最高信用认证（99分），业务覆盖全球100多个国家。总部位于德国，在日本、美国、加拿大、香港等地设有分公司。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + }, + { + "query": "小红产品有什么功效", + "original_query": "小红产品有什么功效", + "rewritten_query": "小红产品有什么功效", + "results": [ + { + "title": "高频问题快速回答", + "content": "FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分，通过NTC营养保送系统直达细胞。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。很多人喝完当天就能感受到精力明显提升。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + }, + { + "query": "大白产品怎么吃", + "original_query": "大白产品怎么吃", + "rewritten_query": "大白产品怎么吃", + "results": [ + { + "title": "高频问题快速回答", + "content": "德国PM大白Basics是基础营养素，为身体提供全面的维生素和矿物质基础。它的核心作用：1.补充每日所需的基础营养。2.为细胞提供全面的营养支持。3.搭配小红和小白形成完整的NTC营养循环。建议早上空腹服用，1平勺兑200-300ml温水。" + } + ], + "total": 1, + "source": "hot_answer_cache", + "hit": true, + "reason": "hot_answer", + "latency_ms": 0, + "hot_answer": true + } + ] + } + } + } + ] +} \ No newline at end of file diff --git a/test2/server/tests/test_viking_cold_start.js b/test2/server/tests/test_viking_cold_start.js new file mode 100644 index 0000000..f3f2d48 --- /dev/null +++ b/test2/server/tests/test_viking_cold_start.js @@ -0,0 +1,91 @@ +const path = require('path'); +const { performance } = require('perf_hooks'); + +require('dotenv').config({ path: path.join(__dirname, '../.env') }); + +const ToolExecutor = require('../services/toolExecutor'); + +const coldTestQueryTemplates = [ + { name: 'Product - Q10 Unique', template: 'Q10辅酵素氧修护有什么独特功效 ' }, + { name: 'Product - IB5 Unique', template: 'IB5口腔免疫喷雾如何正确使用 ' }, + { name: 'Product - CC胶囊 Unique', template: 'CC胶囊的主要适用人群有哪些 ' }, + { name: 'Company - 邓白氏 Unique', template: '德国PM邓白氏认证的具体含义是什么 ' }, + { name: 'Technology - 火炉原理 Unique', template: '请详细阐述一下火炉原理的核心思想 ' } +]; + +async function runColdStartTest() { + console.log('='.repeat(80)); + console.log('VIKING COLD START PERFORMANCE TEST'); + console.log('(No Cache - Real API Calls)'); + console.log('='.repeat(80)); + console.log(''); + + const allResults = []; + + for (const { name, template } of coldTestQueryTemplates) { + console.log(`Testing (Cold): ${name}`); + const latencies = []; + const hits = []; + + for (let i = 0; i < 3; i++) { + const uniqueQuery = template + Math.random(); + const start = performance.now(); + let result; + try { + result = await ToolExecutor.searchKnowledge({ query: uniqueQuery, response_mode: 'answer' }, []); + const latency = performance.now() - start; + latencies.push(latency); + hits.push(!!result.hit); + console.log(` Attempt ${i + 1}: ${latency.toFixed(2)}ms, hit=${result.hit}, source=${result.source}`); + } catch (e) { + console.log(` Attempt ${i + 1} error: ${e.message}`); + } + } + + if (latencies.length > 0) { + const avgLatency = latencies.reduce((a, b) => a + b, 0) / latencies.length; + const minLatency = Math.min(...latencies); + const maxLatency = Math.max(...latencies); + const hitRate = hits.filter(h => h).length / hits.length; + + allResults.push({ + name, + template, + avgLatency, + minLatency, + maxLatency, + hitRate, + latencies + }); + + console.log(` → Avg: ${avgLatency.toFixed(2)}ms, Min: ${minLatency.toFixed(2)}ms, Max: ${maxLatency.toFixed(2)}ms, Hit Rate: ${(hitRate * 100).toFixed(1)}%\n`); + } + } + + console.log('='.repeat(80)); + console.log('COLD START SUMMARY'); + console.log('='.repeat(80)); + + if (allResults.length > 0) { + const totalAvg = allResults.reduce((a, b) => a + b.avgLatency, 0) / allResults.length; + const totalMin = Math.min(...allResults.flatMap(r => r.latencies)); + const totalMax = Math.max(...allResults.flatMap(r => r.latencies)); + const totalHitRate = allResults.reduce((a, b) => a + b.hitRate, 0) / allResults.length; + + console.log(`\nOverall Average Latency: ${totalAvg.toFixed(2)}ms`); + console.log(`Overall Min Latency: ${totalMin.toFixed(2)}ms`); + console.log(`Overall Max Latency: ${totalMax.toFixed(2)}ms`); + console.log(`Overall Hit Rate: ${(totalHitRate * 100).toFixed(1)}%`); + + console.log('\nFastest queries:'); + allResults.sort((a, b) => a.avgLatency - b.avgLatency).forEach((r, i) => { + console.log(` ${i + 1}. ${r.name}: ${r.avgLatency.toFixed(2)}ms`); + }); + } + + console.log('\n' + '='.repeat(80)); + + return allResults; +} + +runColdStartTest().catch(console.error); diff --git a/test2/server/tests/test_viking_direct_api.js b/test2/server/tests/test_viking_direct_api.js new file mode 100644 index 0000000..6bcfefd --- /dev/null +++ b/test2/server/tests/test_viking_direct_api.js @@ -0,0 +1,152 @@ +const path = require('path'); +const { performance } = require('perf_hooks'); +const axios = require('axios'); +const https = require('https'); + +require('dotenv').config({ path: path.join(__dirname, '../.env') }); + +const kbHttpAgent = new https.Agent({ + keepAlive: true, + keepAliveMsecs: 30000, + maxSockets: 6, + timeout: 15000, +}); + +const directTestQueries = [ + { name: 'Q10 Direct', query: 'Q10辅酵素氧修护有什么独特功效' }, + { name: 'IB5 Direct', query: 'IB5口腔免疫喷雾如何正确使用' }, + { name: 'CC胶囊 Direct', query: 'CC胶囊的主要适用人群有哪些' }, + { name: '邓白氏 Direct', query: '德国PM邓白氏认证的具体含义是什么' }, + { name: '火炉原理 Direct', query: '请详细阐述一下火炉原理的核心思想' }, +]; + +async function callVikingDirectly(query, datasetIds) { + const endpointId = process.env.VOLC_ARK_KNOWLEDGE_ENDPOINT_ID || process.env.VOLC_ARK_ENDPOINT_ID; + const authKey = process.env.VOLC_ARK_API_KEY || process.env.VOLC_ACCESS_KEY_ID; + const kbIds = datasetIds || (process.env.VOLC_ARK_KNOWLEDGE_BASE_IDS || '').split(',').map(id => id.trim()).filter(Boolean); + const topK = parseInt(process.env.VOLC_ARK_KNOWLEDGE_TOP_K) || 3; + const threshold = parseFloat(process.env.VOLC_ARK_KNOWLEDGE_THRESHOLD) || 0.4; + + if (!endpointId || !authKey || kbIds.length === 0) { + throw new Error('Missing required config'); + } + + const systemContent = '你是企业知识库问答助手。只依据知识库信息回答，不补充不脑补。'; + + const body = { + model: endpointId, + messages: [ + { role: 'system', content: systemContent }, + { role: 'user', content: query } + ], + metadata: { + knowledge_base: { + dataset_ids: kbIds, + top_k: topK, + threshold: threshold, + }, + }, + stream: false, + max_tokens: 400, + }; + + const response = await axios.post( + 'https://ark.cn-beijing.volces.com/api/v3/chat/completions', + body, + { + headers: { + 'Content-Type': 'application/json', + 'Authorization': `Bearer ${authKey}`, + }, + timeout: 30000, + httpsAgent: kbHttpAgent, + } + ); + + return response.data; +} + +async function runDirectAPITest() { + console.log('='.repeat(80)); + console.log('VIKING DIRECT API PERFORMANCE TEST'); + console.log('(No Query Rewrite, No Cache - Pure API Calls)'); + console.log('='.repeat(80)); + console.log(''); + + const allResults = []; + + for (const { name, query } of directTestQueries) { + console.log(`Testing (Direct): ${name}`); + const latencies = []; + + for (let i = 0; i < 3; i++) { + const uniqueSuffix = ` [${Date.now()}-${Math.random()}]`; + const uniqueQuery = query + uniqueSuffix; + + const start = performance.now(); + try { + const result = await callVikingDirectly(uniqueQuery); + const latency = performance.now() - start; + latencies.push(latency); + + const content = result?.choices?.[0]?.message?.content || 'N/A'; + console.log(` Attempt ${i + 1}: ${latency.toFixed(2)}ms, content length=${content.length}`); + } catch (e) { + console.log(` Attempt ${i + 1} error: ${e.message}`); + if (e.response) { + console.log(` Status: ${e.response.status}, Data:`, JSON.stringify(e.response.data).substring(0, 200)); + } + } + } + + if (latencies.length > 0) { + const avgLatency = latencies.reduce((a, b) => a + b, 0) / latencies.length; + const minLatency = Math.min(...latencies); + const maxLatency = Math.max(...latencies); + + allResults.push({ + name, + query, + avgLatency, + minLatency, + maxLatency, + latencies + }); + + console.log(` → Avg: ${avgLatency.toFixed(2)}ms, Min: ${minLatency.toFixed(2)}ms, Max: ${maxLatency.toFixed(2)}ms\n`); + } + } + + console.log('='.repeat(80)); + console.log('DIRECT API SUMMARY'); + console.log('='.repeat(80)); + + if (allResults.length > 0) { + const totalAvg = allResults.reduce((a, b) => a + b.avgLatency, 0) / allResults.length; + const totalMin = Math.min(...allResults.flatMap(r => r.latencies)); + const totalMax = Math.max(...allResults.flatMap(r => r.latencies)); + const allLatencies = allResults.flatMap(r => r.latencies); + allLatencies.sort((a, b) => a - b); + const p50Index = Math.ceil(0.5 * allLatencies.length) - 1; + const p95Index = Math.ceil(0.95 * allLatencies.length) - 1; + const p99Index = Math.ceil(0.99 * allLatencies.length) - 1; + + console.log(`\nOverall Average Latency: ${totalAvg.toFixed(2)}ms`); + console.log(`Overall P50 Latency: ${allLatencies[p50Index].toFixed(2)}ms`); + console.log(`Overall P95 Latency: ${allLatencies[p95Index].toFixed(2)}ms`); + console.log(`Overall P99 Latency: ${allLatencies[p99Index].toFixed(2)}ms`); + console.log(`Overall Min Latency: ${totalMin.toFixed(2)}ms`); + console.log(`Overall Max Latency: ${totalMax.toFixed(2)}ms`); + + console.log('\nAll queries sorted by latency:'); + allResults.sort((a, b) => a.avgLatency - b.avgLatency).forEach((r, i) => { + console.log(` ${i + 1}. ${r.name}: ${r.avgLatency.toFixed(2)}ms`); + }); + } + + console.log('\n' + '='.repeat(80)); + + return allResults; +} + +runDirectAPITest().catch(console.error); diff --git a/test2/server/tests/viking_retrieval_performance.js b/test2/server/tests/viking_retrieval_performance.js new file mode 100644 index 0000000..db80ea2 --- /dev/null +++ b/test2/server/tests/viking_retrieval_performance.js @@ -0,0 +1,349 @@ +const fs = require('fs'); +const path = require('path'); +const { performance } = require('perf_hooks'); + +require('dotenv').config({ path: path.join(__dirname, '../.env') }); + +const ToolExecutor = require('../services/toolExecutor'); + +class VikingRetrievalPerformanceTester { + constructor(options = {}) { + this.results = []; + this.outputDir = options.outputDir || path.join(__dirname, 'test_results'); + this.verbose = options.verbose !== false; + this.warmupRuns = options.warmupRuns || 2; + + if (!fs.existsSync(this.outputDir)) { + fs.mkdirSync(this.outputDir, { recursive: true }); + } + } + + log(msg) { + if (this.verbose) { + console.log(`[VikingTest] ${msg}`); + } + } + + async warmup(queries) { + this.log('Warming up...'); + for (let i = 0; i < this.warmupRuns; i++) { + for (const query of queries) { + try { + await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + } catch (e) { + // ignore warmup errors + } + } + } + this.log('Warmup complete'); + } + + async testLatency(testQueries, iterations = 10) { + this.log(`Starting latency test with ${iterations} iterations...`); + + const results = {}; + + for (const { name, query } of testQueries) { + this.log(`Testing query: ${name}`); + const latencies = []; + const hits = []; + + for (let i = 0; i < iterations; i++) { + const start = performance.now(); + let result; + try { + result = await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + const latency = performance.now() - start; + latencies.push(latency); + hits.push(!!result.hit); + this.log(` Iteration ${i + 1}: ${latency.toFixed(2)}ms, hit=${result.hit}`); + } catch (e) { + this.log(` Iteration ${i + 1} error: ${e.message}`); + } + } + + results[name] = { + query, + latencies, + hits, + avgLatency: latencies.length ? latencies.reduce((a, b) => a + b, 0) / latencies.length : 0, + minLatency: latencies.length ? Math.min(...latencies) : 0, + maxLatency: latencies.length ? Math.max(...latencies) : 0, + p50Latency: this.percentile(latencies, 50), + p95Latency: this.percentile(latencies, 95), + p99Latency: this.percentile(latencies, 99), + hitRate: hits.length ? hits.filter(h => h).length / hits.length : 0 + }; + } + + this.results.push({ + type: 'latency', + timestamp: new Date().toISOString(), + iterations, + results + }); + + return results; + } + + async testCacheEfficiency(queries, cacheHitsIterations = 5) { + this.log('Testing cache efficiency...'); + + const results = []; + + for (const { name, query } of queries) { + this.log(`Testing cache for query: ${name}`); + + const firstStart = performance.now(); + const firstResult = await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + const firstLatency = performance.now() - firstStart; + + const cacheLatencies = []; + for (let i = 0; i < cacheHitsIterations; i++) { + const start = performance.now(); + const result = await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + const latency = performance.now() - start; + cacheLatencies.push(latency); + this.log(` Cache hit ${i + 1}: ${latency.toFixed(2)}ms, cache_hit=${!!result.cache_hit}`); + } + + const avgCacheLatency = cacheLatencies.reduce((a, b) => a + b, 0) / cacheLatencies.length; + + results.push({ + name, + query, + firstHitLatency: firstLatency, + cacheHitLatencies, + avgCacheLatency, + speedup: firstLatency / avgCacheLatency, + firstHit: firstResult + }); + } + + this.results.push({ + type: 'cache', + timestamp: new Date().toISOString(), + cacheHitsIterations, + results + }); + + return results; + } + + async testConcurrency(queries, concurrencyLevels = [1, 5, 10, 20]) { + this.log('Testing concurrency...'); + + const results = {}; + + for (const concurrency of concurrencyLevels) { + this.log(`Testing concurrency level: ${concurrency}`); + + const startTime = performance.now(); + const promises = []; + + for (let i = 0; i < concurrency; i++) { + const queryObj = queries[i % queries.length]; + promises.push( + ToolExecutor.searchKnowledge({ query: queryObj.query, response_mode: 'answer' }, []) + ); + } + + const allResults = await Promise.all(promises); + const totalTime = performance.now() - startTime; + + const successCount = allResults.filter(r => r && !r.error).length; + const latencies = allResults.map((r, i) => { + return totalTime / concurrency; + }); + + results[concurrency] = { + concurrency, + totalTime, + throughput: concurrency / (totalTime / 1000), + successRate: successCount / concurrency, + results: allResults + }; + + this.log(` Throughput: ${results[concurrency].throughput.toFixed(2)} req/s`); + this.log(` Success rate: ${(results[concurrency].successRate * 100).toFixed(1)}%`); + } + + this.results.push({ + type: 'concurrency', + timestamp: new Date().toISOString(), + concurrencyLevels, + results + }); + + return results; + } + + async testQueryTypes(queryGroups) { + this.log('Testing different query types...'); + + const results = {}; + + for (const [groupName, queries] of Object.entries(queryGroups)) { + this.log(`Testing group: ${groupName}`); + + const groupResults = []; + + for (const query of queries) { + const start = performance.now(); + const result = await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + const latency = performance.now() - start; + + groupResults.push({ + query, + latency, + hit: !!result.hit, + result + }); + } + + results[groupName] = { + queries: groupResults, + avgLatency: groupResults.reduce((a, b) => a + b.latency, 0) / groupResults.length, + hitRate: groupResults.filter(r => r.hit).length / groupResults.length + }; + } + + this.results.push({ + type: 'query_types', + timestamp: new Date().toISOString(), + results + }); + + return results; + } + + percentile(arr, p) { + if (arr.length === 0) return 0; + const sorted = [...arr].sort((a, b) => a - b); + const index = Math.ceil((p / 100) * sorted.length) - 1; + return sorted[Math.max(0, index)]; + } + + generateReport() { + const report = { + generatedAt: new Date().toISOString(), + summary: {}, + tests: this.results + }; + + for (const test of this.results) { + if (test.type === 'latency') { + report.summary.latency = Object.fromEntries( + Object.entries(test.results).map(([name, data]) => [ + name, + { + avg: data.avgLatency.toFixed(2), + p95: data.p95Latency.toFixed(2), + hitRate: (data.hitRate * 100).toFixed(1) + '%' + } + ]) + ); + } else if (test.type === 'cache') { + report.summary.cache = test.results.map(r => ({ + name: r.name, + speedup: r.speedup.toFixed(2) + 'x' + })); + } else if (test.type === 'concurrency') { + report.summary.concurrency = Object.fromEntries( + Object.entries(test.results).map(([level, data]) => [ + level, + { + throughput: data.throughput.toFixed(2) + ' req/s', + successRate: (data.successRate * 100).toFixed(1) + '%' + } + ]) + ); + } + } + + return report; + } + + saveReport(filename = null) { + const report = this.generateReport(); + const filepath = path.join( + this.outputDir, + filename || `viking_performance_${Date.now()}.json` + ); + fs.writeFileSync(filepath, JSON.stringify(report, null, 2)); + this.log(`Report saved to ${filepath}`); + return filepath; + } + + printSummary() { + console.log('\n' + '='.repeat(80)); + console.log('VIKING RETRIEVAL PERFORMANCE TEST SUMMARY'); + console.log('='.repeat(80)); + + const report = this.generateReport(); + + if (report.summary.latency) { + console.log('\n--- Latency Test ---'); + for (const [name, data] of Object.entries(report.summary.latency)) { + console.log(` ${name}:`); + console.log(` Avg: ${data.avg}ms, P95: ${data.p95}ms, Hit Rate: ${data.hitRate}`); + } + } + + if (report.summary.cache) { + console.log('\n--- Cache Efficiency ---'); + for (const r of report.summary.cache) { + console.log(` ${r.name}: Speedup ${r.speedup}`); + } + } + + if (report.summary.concurrency) { + console.log('\n--- Concurrency Test ---'); + for (const [level, data] of Object.entries(report.summary.concurrency)) { + console.log(` ${level} concurrent:`); + console.log(` Throughput: ${data.throughput}, Success: ${data.successRate}`); + } + } + + console.log('\n' + '='.repeat(80)); + } + + async runFullSuite() { + this.log('Starting full Viking retrieval performance test suite...'); + + const testQueries = [ + { name: 'Product Query - Xiaohong', query: '小红产品有什么功效' }, + { name: 'Product Query - Dabai', query: '大白产品怎么吃' }, + { name: 'Company Info', query: '德国PM公司介绍' }, + { name: 'NTC Technology', query: 'NTC营养保送系统原理' }, + { name: 'Hot Answer', query: '基础三合一怎么吃' }, + { name: 'No Hit Query', query: '今天天气怎么样' } + ]; + + await this.warmup(testQueries.map(q => q.query)); + await this.testLatency(testQueries, 10); + await this.testCacheEfficiency(testQueries.slice(0, 3), 5); + await this.testConcurrency(testQueries.slice(0, 3), [1, 3, 5]); + + const queryGroups = { + product: ['小红有什么功效', '大白怎么吃', '小白的作用'], + company: ['PM公司介绍', '邓白氏认证', '总部在哪里'], + technical: ['NTC原理', '火炉原理', '好转反应'] + }; + await this.testQueryTypes(queryGroups); + + this.printSummary(); + this.saveReport(); + + return this.results; + } +} + +module.exports = VikingRetrievalPerformanceTester; + +if (require.main === module) { + (async () => { + const tester = new VikingRetrievalPerformanceTester(); + await tester.runFullSuite(); + })(); +} diff --git a/test2/server/tests/viking_retrieval_performance_with_mock.js b/test2/server/tests/viking_retrieval_performance_with_mock.js new file mode 100644 index 0000000..c8ce852 --- /dev/null +++ b/test2/server/tests/viking_retrieval_performance_with_mock.js @@ -0,0 +1,354 @@ +const fs = require('fs'); +const path = require('path'); +const { performance } = require('perf_hooks'); + +class MockToolExecutor { + static async searchKnowledge({ query, response_mode }, context) { + await new Promise(resolve => setTimeout(resolve, 100 + Math.random() * 200)); + + const mockResults = { + '小红产品有什么功效': { + hit: true, + content: 'FitLine小红Activize Oxyplus的核心功效是提升细胞能量。它含有维生素B族、C和辅酶Q10等成分。主要作用：1.提升精力和体能。2.促进细胞氧气利用。3.增强免疫力。' + }, + '大白产品怎么吃': { + hit: true, + content: '德国PM大白Basics是基础营养素，早上空腹服用，1平勺兑200-300ml温水。' + }, + '德国PM公司介绍': { + hit: true, + content: '德国PM-International是1993年创立的国际营养品直销企业。核心品牌FitLine专注细胞营养，拥有独家NTC营养保送系统技术。' + }, + 'NTC营养保送系统原理': { + hit: true, + content: 'NTC营养保送系统是PM产品的核心技术优势。它能确保营养素在体内精准保送到细胞层面。' + }, + '基础三合一怎么吃': { + hit: true, + hot_answer: true, + content: '基础三合一这样吃：1.大白Basics：早上空腹，1平勺兑200-300ml温水。2.小红Activize：大白喝完15-30分钟后，兑100-150ml温水。3.小白Restorate：睡前空腹，1平勺兑200ml温水。' + }, + '今天天气怎么样': { + hit: false, + content: '知识库中暂未找到相关信息。' + } + }; + + const result = mockResults[query] || { hit: false, content: '未找到相关信息' }; + + return { + query, + results: [{ title: '模拟知识库结果', content: result.content }], + hit: result.hit, + source: result.hot_answer ? 'hot_answer_cache' : 'mock_knowledge', + hot_answer: result.hot_answer + }; + } +} + +class VikingRetrievalPerformanceTester { + constructor(options = {}) { + this.results = []; + this.outputDir = options.outputDir || path.join(__dirname, 'test_results'); + this.verbose = options.verbose !== false; + this.warmupRuns = options.warmupRuns || 2; + this.mockMode = options.mockMode !== false; + + if (!fs.existsSync(this.outputDir)) { + fs.mkdirSync(this.outputDir, { recursive: true }); + } + } + + log(msg) { + if (this.verbose) { + console.log(`[VikingTest] ${msg}`); + } + } + + async warmup(queries) { + this.log('Warming up...'); + const ToolExecutor = this.mockMode ? MockToolExecutor : require('../services/toolExecutor'); + for (let i = 0; i < this.warmupRuns; i++) { + for (const query of queries) { + try { + await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + } catch (e) { + } + } + } + this.log('Warmup complete'); + } + + async testLatency(testQueries, iterations = 10) { + this.log(`Starting latency test with ${iterations} iterations...`); + const ToolExecutor = this.mockMode ? MockToolExecutor : require('../services/toolExecutor'); + + const results = {}; + + for (const { name, query } of testQueries) { + this.log(`Testing query: ${name}`); + const latencies = []; + const hits = []; + + for (let i = 0; i < iterations; i++) { + const start = performance.now(); + let result; + try { + result = await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + const latency = performance.now() - start; + latencies.push(latency); + hits.push(!!result.hit); + this.log(` Iteration ${i + 1}: ${latency.toFixed(2)}ms, hit=${result.hit}`); + } catch (e) { + this.log(` Iteration ${i + 1} error: ${e.message}`); + } + } + + results[name] = { + query, + latencies, + hits, + avgLatency: latencies.length ? latencies.reduce((a, b) => a + b, 0) / latencies.length : 0, + minLatency: latencies.length ? Math.min(...latencies) : 0, + maxLatency: latencies.length ? Math.max(...latencies) : 0, + p50Latency: this.percentile(latencies, 50), + p95Latency: this.percentile(latencies, 95), + p99Latency: this.percentile(latencies, 99), + hitRate: hits.length ? hits.filter(h => h).length / hits.length : 0 + }; + } + + this.results.push({ + type: 'latency', + timestamp: new Date().toISOString(), + iterations, + mockMode: this.mockMode, + results + }); + + return results; + } + + async testCacheEfficiency(queries, cacheHitsIterations = 5) { + this.log('Testing cache efficiency...'); + const ToolExecutor = this.mockMode ? MockToolExecutor : require('../services/toolExecutor'); + + const results = []; + + for (const { name, query } of queries) { + this.log(`Testing cache for query: ${name}`); + + const firstStart = performance.now(); + const firstResult = await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + const firstLatency = performance.now() - firstStart; + + const cacheLatencies = []; + for (let i = 0; i < cacheHitsIterations; i++) { + const start = performance.now(); + const result = await ToolExecutor.searchKnowledge({ query, response_mode: 'answer' }, []); + const latency = performance.now() - start; + cacheLatencies.push(latency); + this.log(` Cache hit ${i + 1}: ${latency.toFixed(2)}ms`); + } + + const avgCacheLatency = cacheLatencies.reduce((a, b) => a + b, 0) / cacheLatencies.length; + + results.push({ + name, + query, + firstHitLatency: firstLatency, + cacheHitLatencies: cacheLatencies, + avgCacheLatency, + speedup: firstLatency / avgCacheLatency, + firstHit: firstResult + }); + } + + this.results.push({ + type: 'cache', + timestamp: new Date().toISOString(), + cacheHitsIterations, + mockMode: this.mockMode, + results + }); + + return results; + } + + async testConcurrency(queries, concurrencyLevels = [1, 5, 10, 20]) { + this.log('Testing concurrency...'); + const ToolExecutor = this.mockMode ? MockToolExecutor : require('../services/toolExecutor'); + + const results = {}; + + for (const concurrency of concurrencyLevels) { + this.log(`Testing concurrency level: ${concurrency}`); + + const startTime = performance.now(); + const promises = []; + + for (let i = 0; i < concurrency; i++) { + const queryObj = queries[i % queries.length]; + promises.push( + ToolExecutor.searchKnowledge({ query: queryObj.query, response_mode: 'answer' }, []) + ); + } + + const allResults = await Promise.all(promises); + const totalTime = performance.now() - startTime; + + const successCount = allResults.filter(r => r && !r.error).length; + + results[concurrency] = { + concurrency, + totalTime, + throughput: concurrency / (totalTime / 1000), + successRate: successCount / concurrency, + results: allResults + }; + + this.log(` Throughput: ${results[concurrency].throughput.toFixed(2)} req/s`); + this.log(` Success rate: ${(results[concurrency].successRate * 100).toFixed(1)}%`); + } + + this.results.push({ + type: 'concurrency', + timestamp: new Date().toISOString(), + concurrencyLevels, + mockMode: this.mockMode, + results + }); + + return results; + } + + percentile(arr, p) { + if (arr.length === 0) return 0; + const sorted = [...arr].sort((a, b) => a - b); + const index = Math.ceil((p / 100) * sorted.length) - 1; + return sorted[Math.max(0, index)]; + } + + generateReport() { + const report = { + generatedAt: new Date().toISOString(), + mockMode: this.mockMode, + summary: {}, + tests: this.results + }; + + for (const test of this.results) { + if (test.type === 'latency') { + report.summary.latency = Object.fromEntries( + Object.entries(test.results).map(([name, data]) => [ + name, + { + avg: data.avgLatency.toFixed(2), + p95: data.p95Latency.toFixed(2), + hitRate: (data.hitRate * 100).toFixed(1) + '%' + } + ]) + ); + } else if (test.type === 'cache') { + report.summary.cache = test.results.map(r => ({ + name: r.name, + speedup: r.speedup.toFixed(2) + 'x' + })); + } else if (test.type === 'concurrency') { + report.summary.concurrency = Object.fromEntries( + Object.entries(test.results).map(([level, data]) => [ + level, + { + throughput: data.throughput.toFixed(2) + ' req/s', + successRate: (data.successRate * 100).toFixed(1) + '%' + } + ]) + ); + } + } + + return report; + } + + saveReport(filename = null) { + const report = this.generateReport(); + const filepath = path.join( + this.outputDir, + filename || `viking_performance_${Date.now()}.json` + ); + fs.writeFileSync(filepath, JSON.stringify(report, null, 2)); + this.log(`Report saved to ${filepath}`); + return filepath; + } + + printSummary() { + console.log('\n' + '='.repeat(80)); + console.log('VIKING RETRIEVAL PERFORMANCE TEST SUMMARY'); + if (this.mockMode) { + console.log('(Mock Mode - For Framework Validation)'); + } + console.log('='.repeat(80)); + + const report = this.generateReport(); + + if (report.summary.latency) { + console.log('\n--- Latency Test ---'); + for (const [name, data] of Object.entries(report.summary.latency)) { + console.log(` ${name}:`); + console.log(` Avg: ${data.avg}ms, P95: ${data.p95}ms, Hit Rate: ${data.hitRate}`); + } + } + + if (report.summary.cache) { + console.log('\n--- Cache Efficiency ---'); + for (const r of report.summary.cache) { + console.log(` ${r.name}: Speedup ${r.speedup}`); + } + } + + if (report.summary.concurrency) { + console.log('\n--- Concurrency Test ---'); + for (const [level, data] of Object.entries(report.summary.concurrency)) { + console.log(` ${level} concurrent:`); + console.log(` Throughput: ${data.throughput}, Success: ${data.successRate}`); + } + } + + console.log('\n' + '='.repeat(80)); + } + + async runFullSuite() { + this.log('Starting full Viking retrieval performance test suite...'); + + const testQueries = [ + { name: 'Product Query - Xiaohong', query: '小红产品有什么功效' }, + { name: 'Product Query - Dabai', query: '大白产品怎么吃' }, + { name: 'Company Info', query: '德国PM公司介绍' }, + { name: 'NTC Technology', query: 'NTC营养保送系统原理' }, + { name: 'Hot Answer', query: '基础三合一怎么吃' }, + { name: 'No Hit Query', query: '今天天气怎么样' } + ]; + + await this.warmup(testQueries.map(q => q.query)); + await this.testLatency(testQueries, 10); + await this.testCacheEfficiency(testQueries.slice(0, 3), 5); + await this.testConcurrency(testQueries.slice(0, 3), [1, 3, 5]); + + this.printSummary(); + this.saveReport(); + + return this.results; + } +} + +module.exports = VikingRetrievalPerformanceTester; + +if (require.main === module) { + (async () => { + const args = process.argv.slice(2); + const mockMode = args[0] !== 'real'; + + const tester = new VikingRetrievalPerformanceTester({ mockMode }); + await tester.runFullSuite(); + })(); +}