AIclinicalresearch/docs/03-业务模块/ASL-AI智能文献/05-开发记录/2025-11-21-字段映射问题修复.md

字段映射问题修å¤<EFBFBD>报告

日期: 2025-11-21
问题: 真实LLM筛选失败（æˆ<C3A6>功ï¼?/20ï¼? **原因**: 字段å<C2B5><C3A5>ä¸<C3A4>匹é…<C3A9>
**状�*: �已修�

---

ðŸ”<EFBFBD> 问题诊断

症状

任务状� completed
进度: 20/20
æˆ<C3A6>功: 0 â<>?
筛选结果数: 0

**表现**�

• 任务瞬间完æˆ<EFBFBD>ï¼?秒）
• 所有文献处ç<EFBFBD>†å¤±è´?
• 没有ä¿<EFBFBD>存任何筛选结æž?

---

🎯 根本原因

问题1: PICOS字段å<C2B5><C3A5>ä¸<C3A4>匹é…<C3A9>

*å‰<EFBFBD>端/æ•°æ<C2B0>®åº“æ ¼å¼? (TitleScreeningSettings.tsx):

picoCriteria: {
  P: '2型糖尿病患�..',
  I: 'SGLT2抑制�..',
  C: '安慰剂或常规治疗...',
  O: '心血管结局...',
  S: 'RCT'
}

LLMæœ<EFBFBD>务期望格å¼<EFBFBD> (llmScreeningService.ts):

// 实际上支æŒ<C3A6>两ç§<C3A7>æ ¼å¼<C3A5>，但优先使用短格å¼<C3A5>
picoCriteria: {
  P: '...',  // �
  I: '...',  // �
  C: '...',  // �
  O: '...',  // �
  S: '...'   // �
}

诊断：å‰<EFBFBD>端使ç”?P/I/C/O/S æ ¼å¼<C3A5>，但 screeningService.ts 直接传递了数æ<C2B0>®åº“的原始格å¼<C3A5>，未å<C2AA>šæ˜ å°„ã€?

---

问题2: 模型å<E280B9><C3A5>æ ¼å¼<C3A5>ä¸<C3A4>匹é…<C3A9>

å‰<EFBFBD>端格å¼<EFBFBD> (TitleScreeningSettings.tsx):

models: ['DeepSeek-V3', 'Qwen-Max']

LLMæœ<EFBFBD>务期望格å¼<EFBFBD> (llmScreeningService.ts):

models: ['deepseek-chat', 'qwen-max']

**原因**：å‰<C3A5>端使用展示å<C2BA><C3A5>称，å<C592>Žç«¯éœ€è¦<C3A8>APIå<49><C3A5>称ã€?

---

问题3: 缺少字段验è¯<C3A8>

文献å<EFBFBD>¯èƒ½ç¼ºå°‘ title æˆ?abstract，导致LLM调用失败ã€?

---

âœ?ä¿®å¤<C3A5>方案

ä¿®å¤<EFBFBD>1: 添加PICOS字段映射

文件: backend/src/modules/asl/services/screeningService.ts

// 🔧 ä¿®å¤<C3A5>：字段å<C2B5><C3A5>映射（数æ<C2B0>®åº“æ ¼å¼<C3A5> â†?LLMæœ<C3A6>务格å¼<C3A5>ï¼?
const rawPicoCriteria = project.picoCriteria as any;
const picoCriteria = {
  P: rawPicoCriteria?.P || rawPicoCriteria?.population || '',
  I: rawPicoCriteria?.I || rawPicoCriteria?.intervention || '',
  C: rawPicoCriteria?.C || rawPicoCriteria?.comparison || '',
  O: rawPicoCriteria?.O || rawPicoCriteria?.outcome || '',
  S: rawPicoCriteria?.S || rawPicoCriteria?.studyDesign || '',
};

**优势**�

• âœ?兼容两ç§<C3A7>æ ¼å¼<C3A5>（P/I/C/O/S æˆ?population/intervention/...ï¼?
• âœ?防御性编程，é<C592>¿å…<C3A5>undefined

---

ä¿®å¤<EFBFBD>2: 添加模型å<E280B9><C3A5>映å°?

// 🔧 ä¿®å¤<C3A5>：模型å<E280B9><C3A5>映射（å‰<C3A5>端格å¼?â†?APIæ ¼å¼<C3A5>ï¼?
const MODEL_NAME_MAP: Record<string, string> = {
  'DeepSeek-V3': 'deepseek-chat',
  'Qwen-Max': 'qwen-max',
  'GPT-4o': 'gpt-4o',
  'Claude-4.5': 'claude-sonnet-4.5',
  'deepseek-chat': 'deepseek-chat',  // 兼容直接使用APIå<49>?
  'qwen-max': 'qwen-max',
  // ... 更多映射
};

const rawModels = screeningConfig?.models || ['deepseek-chat', 'qwen-max'];
const models = rawModels.map((m: string) => MODEL_NAME_MAP[m] || m);

**映射�*�

å‰<EFBFBD>端展示å<EFBFBD>?	APIå<EFBFBD><EFBFBD>称
DeepSeek-V3	deepseek-chat
Qwen-Max	qwen-max
GPT-4o	gpt-4o
Claude-4.5	claude-sonnet-4.5

---

ä¿®å¤<EFBFBD>3: 添加文献验è¯<C3A8>

// 🔧 验è¯<C3A8>：必须有标题和摘è¦?
if (!literature.title || !literature.abstract) {
  logger.warn('Skipping literature without title or abstract', {
    literatureId: literature.id,
    hasTitle: !!literature.title,
    hasAbstract: !!literature.abstract,
  });
  console.log(`âš ï¸<C3AF>  跳过文献 ${processedCount + 1}: 缺少标题或摘è¦<C3A8>`);
  processedCount++;
  continue;
}

---

ä¿®å¤<EFBFBD>4: 增强调试日志

console.log('\n🚀 开始真实LLM筛�');
console.log('  任务ID:', taskId);
console.log('  项目ID:', projectId);
console.log('  文献�', literatures.length);
console.log('  模型（映射å<E2809E>Žï¼?', models);  // â­?显示映射å<E2809E>Žçš„å€?
console.log('  PICOS-P:', picoCriteria.P?.substring(0, 50) || '(ç©?');
console.log('  PICOS-I:', picoCriteria.I?.substring(0, 50) || '(ç©?');
console.log('  PICOS-C:', picoCriteria.C?.substring(0, 50) || '(ç©?');
console.log('  纳入标准:', inclusionCriteria?.substring(0, 50) || '(�');
console.log('  排除标准:', exclusionCriteria?.substring(0, 50) || '(�');

---

🧪 测试步骤

1. é‡<C3A9>å<EFBFBD>¯å<C2AF>Žç«¯ï¼ˆå¿…é¡»ï¼<C3AF>ï¼?

# å<>œæ­¢å½“å‰<C3A5>å<EFBFBD>Žç«¯ï¼ˆCtrl+Cï¼?
cd D:\MyCursor\AIclinicalresearch\backend
npm run dev

2. 测试（å°<C3A5>规模ï¼?

1. 访问å‰<EFBFBD>端
2. 填写PICOS
3. **上传5篇文çŒ?*（先测试å°<C3A5>规模）
4. 点击"开始AIåˆ<C3A5>ç­›"

3. 查看å<E280B9>Žç«¯æŽ§åˆ¶å<C2B6>?

**应该看到**�

🚀 开始真实LLM筛�
  任务ID: xxx
  文献� 5
  模型（映射å<E2809E>Žï¼? [ 'deepseek-chat', 'qwen-max' ]
  PICOS-P: 2型糖尿病患�..
  PICOS-I: SGLT2抑制�..
  PICOS-C: 安慰�..
  纳入标准: æˆ<C3A6>人2åž‹ç³–å°¿ç—…...
  排除标准: 综述ã€<C3A3>系统评ä»?..

[等待10-20秒]

âœ?文献 1/5 处ç<E2809E>†æˆ<C3A6>功
  DS: include / Qwen: include
  冲çª<C3A7>: å<>?

[等待10-20秒]

âœ?文献 2/5 处ç<E2809E>†æˆ<C3A6>功
  DS: exclude / Qwen: exclude
  冲çª<C3A7>: å<>?
...

---

📊 预期效果

ä¿®å¤<EFBFBD>å‰?

• â<EFBFBD>±ï¸<EFBFBD> 1秒完æˆ?0ç¯?
• â<EFBFBD>?æˆ<C3A6>功ï¼?
• â<EFBFBD>?筛选结果数ï¼?

ä¿®å¤<EFBFBD>å<EFBFBD>?

• â<EFBFBD>±ï¸<EFBFBD> 50-100秒完æˆ?篇（æ¯<C3A6>篇10-20秒）
• âœ?æˆ<C3A6>功ï¼?
• âœ?筛选结果数ï¼?
• âœ?è¯<C3A8>æ<EFBFBD>®åŒ…å<E280A6>«çœŸå®žçš„AI分æž<C3A6>
• âœ?è¯<C3A8>æ<EFBFBD>®ä¸<C3A4>包å<E280A6>?模拟è¯<C3A8>æ<EFBFBD>®"

---

🔧 修改文件

• âœ?backend/src/modules/asl/services/screeningService.ts
  • 添加PICOS字段映射
  • 添加模型å<EFBFBD><EFBFBD>映å°?
  • 添加文献验è¯<EFBFBD>
  • 增强调试日志

---

💡 ç»<C3A7>验教训

1. å‰<C3A5>å<EFBFBD>Žç«¯æ•°æ<C2B0>®æ ¼å¼<C3A5>一致æ€?

• å‰<EFBFBD>端使用的展示格å¼?â‰?å<>Žç«¯APIæ ¼å¼<C3A5>
• 需è¦<EFBFBD>在集æˆ<EFBFBD>层å<EFBFBD>šæ˜ å°„

2. 防御性编�

• 使用 || æ<><C3A6>供默认å€?
• 验è¯<EFBFBD>必需字段
• 兼容多ç§<EFBFBD>æ ¼å¼<EFBFBD>

3. 调试日志的é‡<C3A9>è¦<C3A8>æ€?

• 显示映射å<EFBFBD>Žçš„值（ä¸<EFBFBD>是原始值）
• 输出所有关键å<EFBFBD>‚æ•?
• 帮助快速定ä½<EFBFBD>é—®é¢?

---

🎯 å<>Žç»­ä¼˜åŒ–

短期

1. âœ?字段映射（已完æˆ<C3A6>ï¼?
2. âœ?模型å<E280B9><C3A5>映射（已完æˆ<C3A6>）
3. âœ?验è¯<C3A8>必需字段（已完æˆ<C3A6>ï¼?

中期

1. 统一å‰<EFBFBD>å<EFBFBD>Žç«¯æ•°æ<EFBFBD>®æ ¼å¼<EFBFBD>（使用 TypeScript 接å<C2A5>£ï¼?
2. 添加数æ<EFBFBD>®æ ¼å¼<EFBFBD>验è¯<EFBFBD>中间ä»?
3. 改进错误æ<EFBFBD><EFBFBD>示

长期

1. 使用 tRPC æˆ?GraphQL ç¡®ä¿<C3A4>类型安全
2. 自动化测试覆�
3. Schema验è¯<EFBFBD>

---

**报告�*: AI Assistant
日期: 2025-11-21
版本: v1.0.0