HaHafeng
1b53ab9d52
Major Changes: - Add StreamingService with OpenAI Compatible format - Upgrade Chat component V2 with Ant Design X integration - Implement AIA module with 12 intelligent agents - Update API routes to unified /api/v1 prefix - Update system documentation Backend (~1300 lines): - common/streaming: OpenAI Compatible adapter - modules/aia: 12 agents, conversation service, streaming integration - Update route versions (RVW, PKB to v1) Frontend (~3500 lines): - modules/aia: AgentHub + ChatWorkspace (100% prototype restoration) - shared/Chat: AIStreamChat, ThinkingBlock, useAIStream Hook - Update API endpoints to v1 Documentation: - AIA module status guide - Universal capabilities catalog - System overview updates - All module documentation sync Tested: Stream response verified, authentication working Status: AIA V2.0 core completed (85%)
1507 lines
42 KiB
Markdown
1507 lines
42 KiB
Markdown
# 蜈ｨ譁<EFBDA8>､咲ｭ幄ｴｨ驥丈ｿ晞囿荳主庄霑ｽ貅ｯ遲也払
|
||
|
||
> **譁<>｡｣迚域悽<E59F9F>?* V1.0
|
||
> **蛻帛ｻｺ譌･譛滂ｼ?* 2025-11-22
|
||
> **騾ら畑讓｡蝮暦ｼ?* AI 譎ｺ閭ｽ譁<EFBDBD>鍵 - 蜈ｨ譁<EFBDA8>､咲ｭ<E592B2>
|
||
> **逶ｮ譬<EFBDAE>ｼ?* 蛻<>亳谿ｵ謠仙合蜈ｨ譁<EFBDA8>､咲ｭ帷噪蜃<E599AA>｡ｮ邇<EFBDAE>∵婿豕募ｭｦ雍ｨ驥丞愛譁ｭ蜥悟ｮ梧紛蜿ｯ霑ｽ貅ｯ諤?
|
||
|
||
---
|
||
|
||
## <20>搭 譁<>｡｣讎りｿｰ
|
||
|
||
譛ｬ譁<EFBFBD>｡｣螳壻ｹ我ｺ<EFBFBD>**蜈ｨ譁<EFBDA8>､咲ｭ帶ｨ｡蝮<EFBDA1>**蝨?**MVP 竊?V1.0 竊?V2.0** 荳我ｸｪ髦ｶ谿ｵ逧<EFBDB5>ｴｨ驥丈ｿ晞囿遲也払縲?
|
||
|
||
### 蜈ｨ譁<EFBDA8>､咲ｭ<E592B2> vs 譬<>｢俶遭隕∝<E99A95>遲幢ｼ壽<EFBDBC>ｸ蠢<EFBDB8>ｷｮ蠑?
|
||
|
||
| 扈ｴ蠎ｦ | 譬<>｢俶遭隕∝<E99A95>遲<EFBFBD> | 蜈ｨ譁<EFBDA8>､咲ｭ<E592B2> | 遲也払蟾ｮ蠑<EFBDAE> |
|
||
|------|-------------|---------|---------|
|
||
| **菫｡諱ｯ驥?* | 200-500蟄?| 5,000-20,000蟄?| <20>閥 髴蛻<C280>ｮｵ螟<EFBDB5>炊 |
|
||
| **蛻､譁ｭ萓晄紺** | PICOS蛹ｹ驟榊ｺ?| 12蟄玲ｮｵ譁ｹ豕募ｭｦ雍ｨ驥?| <20>閥 髴荳謎ｸ壼愛譁ｭ譬<EFBDAD>㊥ |
|
||
| **蜀ｳ遲門､肴揩蠎?* | 菴?譏?蜷? | 鬮?12荳ｪ蟄玲ｮｵﾃ?郤? | <20>閥 髴扈捺桷蛹匁署蜿?|
|
||
| **螳ｹ髞咏ｭ也払** | 螳<>漠蜍ｿ貍<EFBDBF> | 荳崎<E88DB3>貍丞<E8B28D>髞ｮ菫｡諱?| <20>閥 髴鬪瑚ｯ∵惻蛻ｶ |
|
||
| **Token謌先悽** | ﾂ･0.005/遽?| ﾂ･0.05-0.20/遽?| <20>閥 髴謌先悽莨伜喧 |
|
||
| **蜿ｯ霑ｽ貅ｯ諤?* | 蠑慕畑鞫倩ｦ<E580A9> | 蜈ｷ菴馴｡ｵ遐<EFBDB5>/谿ｵ關ｽ/陦ｨ譬ｼ | <20>閥 髴隸∵紺體?|
|
||
|
||
### 譬ｸ蠢<EFBDB8>ｮｾ隶｡蜴溷<E89CB4>
|
||
|
||
| 蜴溷<E89CB4> | 隸ｴ譏<EFBDB4> |
|
||
|------|------|
|
||
| **蠕ｪ隸∝現蟄ｦ譬<EFBDA6>㊥** | 蝓ｺ莠擦ochrane RoB 2.0蟾･蜈ｷ逧<EFBFBD>婿豕募ｭｦ雍ｨ驥剰ｯ<EFBFBD>ｼｰ譬<EFBFBD>㊥ |
|
||
| **扈捺桷蛹匁署蜿?* | Nougat + 蛻<>ｮｵ謠仙叙 + 蜈ｨ譁<EFBDA8>ｪ瑚ｯ<E7919A>ｼ碁∩蜈?Lost in the Middle" |
|
||
| **螳梧紛隸∵紺體?* | 豈丈ｸｪ蟄玲ｮｵ蠑ｺ蛻ｶ隕∵ｱょ次譁<E6ACA1>ｼ慕畑<E68595>磯｡ｵ遐√∵ｮｵ關ｽ縲∬｡ｨ譬ｼ<E8ADAC><EFBDBC> |
|
||
| **蛻<>ｭ･螳樊命** | MVP蜈磯ｪ瑚ｯ∝庄陦梧ｧ<C280>祁1.0謠仙合雍ｨ驥擾ｼ祁2.0霎ｾ蛻ｰ蛹ｻ蟄ｦ郤ｧ譬<EFBFBD><EFBFBD>?|
|
||
| **謌先悽荳手ｴｨ驥丞ｹｳ陦?* | MVP逕ｨ謌先悽蜿句･ｽ讓｡蝙具ｼ悟<EFBDBC>髞ｮ蟄玲ｮｵ逕ｨ鬮倡ｫｯ讓｡蝙矩ｪ瑚ｯ?|
|
||
|
||
---
|
||
|
||
## <20>識 荳蛾亳谿ｵ霍ｯ郤ｿ蝗ｾ
|
||
|
||
```
|
||
MVP (3蜻? V1.0 (5蜻? V2.0 (8蜻?
|
||
笏懌楳 Nougat扈捺桷蛹匁署蜿? 笏懌楳 Cochrane譬<65>㊥Prompt 笏懌楳 荳画ｨ｡蝙倶ｻｲ陬?
|
||
笏懌楳 12蟄玲ｮｵ蛻<EFBDB5>ｮｵ謠仙叙 笏懌楳 Few-shot蛹ｻ蟄ｦ譯井ｾ句ｺ? 笏懌楳 蛹ｻ蟄ｦ騾ｻ霎題ｧ<E9A18C><EFBDA7>蠑墓梼
|
||
笏懌楳 蜿梧ｨ｡蝙矩ｪ瑚ｯ? 笏懌楳 螳梧紛隸∵紺體? 笏懌楳 閾ｪ蜉ｨ雍ｨ驥丞ｮ｡隶｡
|
||
笏懌楳 蟄玲ｮｵ郤ｧ蜀ｲ遯∵｣豬? 笏懌楳 蜈ｨ譁<EFBDA8>ｺ､蜿蛾ｪ瑚ｯ<E7919A> 笏懌楳 HITL譎ｺ閭ｽ蛻<EFBDBD>ｵ<EFBFBD>
|
||
笏披楳 蝓ｺ遑蜿ｯ霑ｽ貅? 笏披楳 蛻<>ｺｧ莠ｺ蟾･螟肴<E89E9F>ｸ 笏披楳 螳｡隶｡郤ｧ譌･蠢?
|
||
竊? 竊? 竊?
|
||
蜃<>｡ｮ邇?竕?85% 蜃<>｡ｮ邇?竕?92% 蜃<>｡ｮ邇?竕?96%
|
||
```
|
||
|
||
---
|
||
|
||
## <20>噫 MVP 髦ｶ谿ｵ<E8B0BF>? 蜻ｨ<E89CBB><EFBDA8>
|
||
|
||
### 逶ｮ譬<EFBDAE>ｮ壻ｽ<E5A3BB>
|
||
|
||
- **蜃<>｡ｮ邇<EFBDAE>岼譬?*<2A>壺翁 85%
|
||
- **菫｡諱ｯ螳梧紛邇?*<2A>壺翁 90%<25>?2蟄玲ｮｵ荳埼＠貍擾ｼ<E693BE>
|
||
- **謌先悽鬚<E682BD>ｮ<EFBFBD>**<2A>壺王 ﾂ･0.05/遽<>ｼ<EFBFBD>eepSeek-V3 + Qwen3-Max<61>?
|
||
- **莠､莉俶<E88E89><E4BFB6>㊥**<2A>壼渕遑蜉溯<E89C89>蜿ｯ逕ｨ<E98095>梧髪謖∫ｻ捺桷蛹匁署蜿門柱蜿梧ｨ｡蝙矩ｪ瑚ｯ<E7919A>
|
||
|
||
---
|
||
|
||
### 荳縲∵<E7B8B2>ｸ蠢<EFBDB8>橿譛ｯ遲也<E981B2>?
|
||
|
||
#### 1.1 笨?Nougat扈捺桷蛹匁署蜿厄ｼ亥<EFBDBC>髞ｮ莨伜漢<E4BC9C>?
|
||
|
||
**荳ｺ莉荵磯画叫Nougat**<EFBFBD>?
|
||
|
||
| 蟇ｹ豈皮ｻｴ蠎ｦ | PyMuPDF | Nougat |
|
||
|---------|---------|--------|
|
||
| 霎灘<E99C8E>譬ｼ蠑<EFBDBC> | 郤ｯ譁<EFBDAF><E8AD81>?| Markdown扈捺桷蛹?|
|
||
| 遶<>闃りｯ<E3828A>悪 | 髴LLM莠梧ｬ｡隸<EFBDA1>悪<EFBFBD>?0%蜃<>｡ｮ邇<EFBDAE>ｼ<EFBFBD> | 螟ｩ辟ｶ菫晉蕗扈捺桷<E68DBA>?5%蜃<>｡ｮ邇<EFBDAE>ｼ俄<EFBDBC>?|
|
||
| 陦ｨ譬ｼ螟<EFBDBC>炊 | 譁<>悽荵ｱ遐<EFBDB1> | Markdown陦ｨ譬ｼ 笨?|
|
||
| 蜈ｬ蠑剰ｯ<E589B0>悪 | 荵ｱ遐<EFBDB1> | LaTeX譬ｼ蠑<EFBDBC> 笨?|
|
||
| 騾ら畑蝨ｺ譎ｯ | 荳ｭ譁<EFBDAD>ｮｺ譁<EFBDBA> | 闍ｱ譁<EFBDB1>ｭｦ譛ｯ隶ｺ譁<EFBDBA> 笨?|
|
||
|
||
**螳樊命譁ｹ譯<EFBDB9>**<EFBFBD>?
|
||
|
||
```typescript
|
||
// 豺ｷ蜷育ｭ也払<E4B99F>哢ougat莨伜<E88EA8><E4BC9C>訓yMuPDF髯咲ｺｧ
|
||
async function extractFullText(pdfBuffer: Buffer, filename: string) {
|
||
// Step 1: 譽豬玖ｯｭ險
|
||
const language = await detectLanguage(pdfBuffer);
|
||
|
||
// Step 2: 闍ｱ譁<EFBDB1>ｮｺ譁<EFBDBA>ｼ伜<EFBDBC>逕ｨNougat
|
||
if (language === 'english') {
|
||
try {
|
||
const nougatResult = await extractionClient.extractPdf(
|
||
pdfBuffer, filename, 'nougat'
|
||
);
|
||
|
||
if (nougatResult.quality > 0.8) {
|
||
return {
|
||
method: 'nougat',
|
||
text: nougatResult.text,
|
||
format: 'markdown',
|
||
structured: true // 箝?蜈ｳ髞ｮ莨伜漢
|
||
};
|
||
}
|
||
} catch (error) {
|
||
console.warn('Nougat螟ｱ雍･<E99B8D>碁剄郤ｧ蛻ｰPyMuPDF');
|
||
}
|
||
}
|
||
|
||
// Step 3: 荳ｭ譁<EFBDAD>ｮｺ譁<EFBDBA><E8AD81>Nougat螟ｱ雍･<E99B8D>檎畑PyMuPDF
|
||
const pymupdfResult = await extractionClient.extractPdf(
|
||
pdfBuffer, filename, 'pymupdf'
|
||
);
|
||
|
||
return {
|
||
method: 'pymupdf',
|
||
text: pymupdfResult.text,
|
||
format: 'plaintext',
|
||
structured: false // 髴隕´LM隸<4D>悪扈捺桷
|
||
};
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
#### 1.2 笨?12蟄玲ｮｵ蛻<EFBDB5>ｮｵ謠仙叙<E4BB99>磯∩蜈広ost in the Middle<6C>?
|
||
|
||
**譬ｸ蠢<EFBDB8>琉鬚<E79089>**<EFBFBD>壼<EFBFBD>譁?0K tokens荳谺｡諤ｧ蝟らｻ儉LM<4C>御ｸｭ髣ｴ遶<EFBDB4>闃ゆｿ｡諱ｯ驕玲ｼ冗紫鬮倩ｾｾ33%
|
||
|
||
**隗｣蜀ｳ譁ｹ譯<EFBDB9>**<EFBFBD>壽潔蟄玲ｮｵ螳壼髄謠仙叙逶ｸ蜈ｳ遶<EFBFBD>闃<EFBFBD>
|
||
|
||
```typescript
|
||
// 12蟄玲ｮｵ謠仙叙霍ｯ逕ｱ陦?
|
||
const FIELD_EXTRACTION_ROUTES = {
|
||
'遐皮ｩｶ隶ｾ隶｡': {
|
||
sections: ['abstract', 'methods'],
|
||
maxTokens: 3000,
|
||
priority: 'high'
|
||
},
|
||
'遐皮ｩｶ莠ｺ鄒､': {
|
||
sections: ['methods', 'results'],
|
||
maxTokens: 3500,
|
||
priority: 'high',
|
||
lookForTables: true // Table 1: Baseline
|
||
},
|
||
'蟷ｲ鬚<EFBDB2>蒔譁ｽ': {
|
||
sections: ['methods', 'results'],
|
||
maxTokens: 3000,
|
||
priority: 'high'
|
||
},
|
||
'蟇ｹ辣ｧ謗ｪ譁ｽ': {
|
||
sections: ['methods', 'results'],
|
||
maxTokens: 2500,
|
||
priority: 'high'
|
||
},
|
||
'扈灘ｱ謖<C280><E8AC96><EFBFBD>': {
|
||
sections: ['methods', 'results'],
|
||
maxTokens: 4000,
|
||
priority: 'high',
|
||
lookForTables: true // Results tables
|
||
},
|
||
'髫乗惻蛹匁婿豕?: {
|
||
sections: ['methods', 'figures'],
|
||
maxTokens: 2500,
|
||
priority: 'critical', // 蜈ｳ髞ｮ蟄玲ｮｵ
|
||
keywords: ['randomization', 'allocation', 'sequence', 'CONSORT']
|
||
},
|
||
'逶ｲ豕<EFBFBD>': {
|
||
sections: ['methods'],
|
||
maxTokens: 2000,
|
||
priority: 'critical'
|
||
},
|
||
'譬ｷ譛ｬ驥剰ｮ｡邂?: {
|
||
sections: ['methods'],
|
||
maxTokens: 2000,
|
||
priority: 'medium'
|
||
},
|
||
'蝓ｺ郤ｿ蜿ｯ豈疲?: {
|
||
sections: ['results', 'tables'],
|
||
maxTokens: 3000,
|
||
priority: 'high',
|
||
specificTable: 'Table 1'
|
||
},
|
||
'扈捺棡螳梧紛諤?: {
|
||
sections: ['results', 'figures'],
|
||
maxTokens: 4000,
|
||
priority: 'critical',
|
||
keywords: ['ITT', 'per-protocol', 'missing data', 'dropout']
|
||
},
|
||
'騾画叫諤ｧ謚･蜻?: {
|
||
sections: ['methods', 'results', 'supplementary'],
|
||
maxTokens: 3000,
|
||
priority: 'medium',
|
||
checkTrialRegistry: true // 蟇ｹ豈疲ｳｨ蜀梧婿譯<E5A9BF>
|
||
},
|
||
'蜈ｶ莉門￥蛟?: {
|
||
sections: ['methods', 'discussion', 'supplementary'],
|
||
maxTokens: 3000,
|
||
priority: 'medium'
|
||
}
|
||
};
|
||
|
||
// 蛻<>ｮｵ蟷ｶ陦梧署蜿<E7BDB2>
|
||
async function extractAllFields(sections: ParsedSections) {
|
||
const extractionTasks = Object.entries(FIELD_EXTRACTION_ROUTES).map(
|
||
([fieldName, config]) => ({
|
||
field: fieldName,
|
||
task: extractFieldWithEvidence(fieldName, sections, config)
|
||
})
|
||
);
|
||
|
||
// 蟷ｶ陦梧鴬陦鯉ｼ磯剄菴主ｻｶ霑滂ｼ<E6BB82>
|
||
const results = await Promise.all(
|
||
extractionTasks.map(t => t.task)
|
||
);
|
||
|
||
return results;
|
||
}
|
||
```
|
||
|
||
**莨伜漢**<EFBFBD>?
|
||
- 笨?驕ｿ蜈堺ｸｭ髣ｴ菫｡諱ｯ驕玲ｼ擾ｼ亥㊥遑ｮ邇<EFBDAE> 70% 竊?90%<25>?
|
||
- 笨?Token豸郁鈴剄菴?0%<25>?0K 竊?12K<32>?
|
||
- 笨?蟷ｶ陦梧署蜿厄ｼ悟ｻｶ霑滄剄菴?0%
|
||
- 笨?豈丈ｸｪ蟄玲ｮｵLLM豕ｨ諢丞鴨譖ｴ髮<EFBDB4>ｸｭ
|
||
|
||
---
|
||
|
||
#### 1.3 笨?蜿梧ｨ｡蝙倶ｺ､蜿蛾ｪ瑚ｯ?
|
||
|
||
**讓｡蝙狗ｻ<E78B97>粋**<EFBFBD>咼eepSeek-V3 + Qwen3-Max<61>域<EFBFBD>譛ｬ蜿句･ｽ<EFBDA5><EFBDBD>
|
||
|
||
```typescript
|
||
// 蜿梧ｨ｡蝙句ｹｶ陦瑚ｰ<E7919A><EFBDB0>?
|
||
async function dualModelExtraction(
|
||
fieldName: string,
|
||
relevantContent: string,
|
||
prompt: string
|
||
) {
|
||
const [resultA, resultB] = await Promise.all([
|
||
llmService.chat('deepseek-v3', prompt, relevantContent),
|
||
llmService.chat('qwen-max', prompt, relevantContent)
|
||
]);
|
||
|
||
// 隗｣譫千ｻ捺棡
|
||
const assessmentA = parseFieldAssessment(resultA);
|
||
const assessmentB = parseFieldAssessment(resultB);
|
||
|
||
// 蜀ｲ遯∵｣豬?
|
||
const hasConflict = assessmentA.level !== assessmentB.level;
|
||
|
||
return {
|
||
field: fieldName,
|
||
modelA: {
|
||
model: 'deepseek-v3',
|
||
assessment: assessmentA.level, // '螳梧紛'/'荳榊ｮ梧<EFBDAE>?/'譌<>豕募愛譁ｭ'
|
||
evidence: assessmentA.evidence,
|
||
confidence: assessmentA.confidence
|
||
},
|
||
modelB: {
|
||
model: 'qwen-max',
|
||
assessment: assessmentB.level,
|
||
evidence: assessmentB.evidence,
|
||
confidence: assessmentB.confidence
|
||
},
|
||
hasConflict,
|
||
needReview: hasConflict ||
|
||
assessmentA.confidence < 0.7 ||
|
||
assessmentB.confidence < 0.7
|
||
};
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
#### 1.4 笨?蟄玲ｮｵ郤ｧ蜀ｲ遯∵｣豬倶ｸ主<EFBDB8>郤ｧ螟肴<E89E9F>ｸ
|
||
|
||
**荳肴弍邂蜊慕噪"蜈ｨ驛ｨ蜀ｲ遯∝ｰｱ莠ｺ蟾･螟肴<E89E9F>?<3F>瑚梧弍譬ｹ謐ｮ蟄玲ｮｵ驥崎ｦ∵ｧ蛻<EFBDA7>ｺ?*<2A>?
|
||
|
||
```typescript
|
||
// 蟄玲ｮｵ驥崎ｦ∵ｧ蛻<EFBDA7>ｺ?
|
||
const FIELD_IMPORTANCE = {
|
||
critical: ['髫乗惻蛹匁婿豕?, '逶ｲ豕<EFBFBD>', '扈捺棡螳梧紛諤?], // 譬ｸ蠢<EFBDB8>￥蛟夐｣朱<EFBDA3>?
|
||
high: ['遐皮ｩｶ隶ｾ隶｡', '遐皮ｩｶ莠ｺ鄒､', '蟷ｲ鬚<EFBDB2>蒔譁ｽ', '扈灘ｱ謖<C280><E8AC96><EFBFBD>', '蝓ｺ郤ｿ蜿ｯ豈疲?],
|
||
medium: ['譬ｷ譛ｬ驥剰ｮ｡邂?, '騾画叫諤ｧ謚･蜻?, '蜈ｶ莉門￥蛟?]
|
||
};
|
||
|
||
// 譎ｺ閭ｽ蛻<EFBDBD>ｵ<EFBFBD>
|
||
function prioritizeReview(conflicts: FieldConflict[]): ReviewQueue {
|
||
const queue = {
|
||
urgent: [], // 蜈ｳ髞ｮ蟄玲ｮｵ蜀ｲ遯<EFBDB2> 竊?遶句叉莠ｺ蟾･螟肴<E89E9F>ｸ
|
||
important: [], // 鬮倅ｼ伜<EFBDBC>郤ｧ蟄玲ｮｵ蜀ｲ遯<EFBDB2> 竊?24蟆乗慮蜀<E685AE>､肴<EFBDA4>?
|
||
normal: [] // 荳ｭ遲我ｼ伜<EFBDBC>郤ｧ蟄玲ｮｵ蜀ｲ遯?竊?48蟆乗慮蜀<E685AE>､肴<EFBDA4>?
|
||
};
|
||
|
||
for (const conflict of conflicts) {
|
||
if (!conflict.hasConflict) continue;
|
||
|
||
if (FIELD_IMPORTANCE.critical.includes(conflict.field)) {
|
||
queue.urgent.push({
|
||
...conflict,
|
||
reason: '蜈ｳ髞ｮ譁ｹ豕募ｭｦ蟄玲ｮｵ蜀ｲ遯<EFBDB2>ｼ悟ｽｱ蜩榊￥蛟夐｣朱勦隸<E58BA6>ｼ?,
|
||
deadline: new Date(Date.now() + 2 * 3600 * 1000) // 2蟆乗慮
|
||
});
|
||
} else if (FIELD_IMPORTANCE.high.includes(conflict.field)) {
|
||
queue.important.push({
|
||
...conflict,
|
||
reason: '鬮倅ｼ伜<EFBFBD>郤ｧ蟄玲ｮｵ蜀ｲ遯<EFBFBD>',
|
||
deadline: new Date(Date.now() + 24 * 3600 * 1000) // 24蟆乗慮
|
||
});
|
||
} else {
|
||
queue.normal.push({
|
||
...conflict,
|
||
reason: '荳闊ｬ蟄玲ｮｵ蜀ｲ遯?,
|
||
deadline: new Date(Date.now() + 48 * 3600 * 1000) // 48蟆乗慮
|
||
});
|
||
}
|
||
}
|
||
|
||
return queue;
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
#### 1.5 笨?蝓ｺ遑隸∵紺體ｾ<E9AB94>亥次譁<E6ACA1>ｼ慕畑<E68595>?
|
||
|
||
**MVP髦ｶ谿ｵ隕∵ｱ<E288B5>**<EFBFBD>壽ｯ丈ｸｪ蟄玲ｮｵ蠢<EFBFBD>｡ｻ譛牙次譁<EFBFBD>ｼ慕畑
|
||
|
||
```typescript
|
||
interface FieldEvidence {
|
||
field: string;
|
||
assessment: '螳梧紛' | '荳榊ｮ梧<EFBDAE>? | '譌<EFBFBD>豕募愛譁ｭ';
|
||
|
||
// 箝?蠑ｺ蛻ｶ隕∵ｱ<E288B5>
|
||
evidence: {
|
||
quote: string; // 蜴滓枚蠑慕畑<E68595>?00-300蟄暦ｼ<E69AA6>
|
||
location: {
|
||
section: string; // "Methods"
|
||
page?: number; // 3<>亥ｦよ棡PDF譛蛾｡ｵ遐<EFBDB5>ｼ<EFBFBD>
|
||
paragraph?: number; // 2
|
||
table?: string; // "Table 1"
|
||
figure?: string; // "Figure 1"
|
||
};
|
||
highlightedKeywords: string[]; // 蜈ｳ髞ｮ菫｡蜿ｷ隸?
|
||
};
|
||
|
||
reasoning: string; // 蛻､譁ｭ逅<EFBDAD>罰<EFBFBD>?0-200蟄暦ｼ<E69AA6>
|
||
confidence: number; // 0.0-1.0
|
||
}
|
||
|
||
// 蜷主､<E4B8BB>炊鬪瑚ｯ<E7919A>ｼ夂｡ｮ菫晄ｯ丈ｸｪ蟄玲ｮｵ驛ｽ譛芽ｯ∵紺
|
||
function validateEvidence(result: ExtractionResult): ValidationReport {
|
||
const errors = [];
|
||
|
||
for (const [field, data] of Object.entries(result.fields)) {
|
||
// 譽譟?<3F>壼ｿ<E5A3BC>｡ｻ譛牙ｼ慕畑
|
||
if (!data.evidence?.quote) {
|
||
errors.push({
|
||
field,
|
||
type: 'missing_evidence',
|
||
message: `蟄玲ｮｵ"${field}"郛ｺ蟆大次譁<E6ACA1>ｼ慕畑`
|
||
});
|
||
}
|
||
|
||
// 譽譟?<3F>壼ｼ慕畑荳崎<E88DB3>螟ｪ遏ｭ<E9818F>磯∩蜈肴聞陦搾ｼ?
|
||
if (data.evidence?.quote && data.evidence.quote.length < 50) {
|
||
errors.push({
|
||
field,
|
||
type: 'insufficient_evidence',
|
||
message: `蟄玲ｮｵ"${field}"逧<>ｼ慕畑霑<E79591>洒<EFBFBD><E6B492><50蟄暦ｼ会ｼ悟庄閭ｽ荳崎ｶｳ莉･謾ｯ謖∝愛譁ｭ`
|
||
});
|
||
}
|
||
|
||
// 譽譟?<3F>壼ｿ<E5A3BC>｡ｻ譛我ｽ咲ｽｮ菫｡諱ｯ
|
||
if (!data.evidence?.location?.section) {
|
||
errors.push({
|
||
field,
|
||
type: 'missing_location',
|
||
message: `蟄玲ｮｵ"${field}"譛ｪ譬<EFBDAA>ｳｨ蜴滓枚菴咲ｽｮ`
|
||
});
|
||
}
|
||
}
|
||
|
||
return {
|
||
isValid: errors.length === 0,
|
||
errors,
|
||
completeness: 1 - (errors.length / (Object.keys(result.fields).length * 3))
|
||
};
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
### 莠後?2蟄玲ｮｵ荳謎ｸ啀rompt讓｡譚ｿ<E8AD9A><EFBDBF>VP迚茨ｼ<E88CA8>
|
||
|
||
#### 遉ｺ萓具ｼ夐囂譛ｺ蛹匁婿豕包ｼ亥<EFBDBC>髞ｮ蟄玲ｮｵ<EFBDAE><EFBDB5>
|
||
|
||
```markdown
|
||
# 蟄玲ｮｵ謠仙叙莉ｻ蜉｡<E89C89>夐囂譛ｺ蛹匁婿豕<E5A9BF>
|
||
|
||
## 閭梧勹隸ｴ譏<EFBDB4>
|
||
菴<EFBFBD>譏ｯ荳菴榊ｾｪ隸∝現蟄ｦ荳灘ｮｶ<EFBFBD>梧ｭ｣蝨ｨ隸<EFBFBD>ｼｰ荳遽⑲CT遐皮ｩｶ逧<EFBFBD>婿豕募ｭｦ雍ｨ驥上?
|
||
隸ｷ譬ｹ謐ｮCochrane蛛丞夐｣朱勦隸<EFBFBD>ｼｰ蟾･蜈ｷ<EFBFBD><EFBFBD>oB 2.0<EFBFBD>臥噪譬<EFBFBD>㊥<EFBFBD>悟愛譁ｭ隸･遐皮ｩｶ逧<EFBFBD>囂譛ｺ蛹匁婿豕墓弍蜷ｦ蜈<EFBFBD><EFBFBD>縲?
|
||
|
||
## 蠕<><E8A095>譫仙<E8ADAB>螳?
|
||
莉･荳区弍隶ｺ譁<EFBFBD>噪Methods遶<EFBFBD>闃ょ柱逶ｸ蜈ｳ蝗ｾ陦ｨ<EFBFBD><EFBFBD>
|
||
|
||
${relevantContent}
|
||
|
||
## 蛻､譁ｭ譬<EFBDAD>㊥
|
||
|
||
### 螳梧紛<E6A2A7><E7B49B>ow risk of bias<61>?
|
||
髴**蜷梧慮貊｡雜ｳ**莉･荳区擅莉ｶ<E88E89>?
|
||
1. 笨?譏守｡ｮ隸ｴ譏朱囂譛ｺ蠎丞<E8A08E>逕滓<E98095>譁ｹ豕<EFBDB9>
|
||
- 遉ｺ萓具ｼ喞omputer-generated random sequence, random number table,
|
||
central randomization, minimization
|
||
2. 笨?隸ｴ譏主<E8AD8F>驟埼嚼阯乗婿豕<E5A9BF>
|
||
- 遉ｺ萓具ｼ嘖ealed opaque envelopes, central allocation,
|
||
pharmacy-controlled, IWRS (Interactive Web Response System)
|
||
3. 笨?譌<>騾画叫蛛丞夂噪隸∵紺
|
||
- 蝓ｺ郤ｿ迚ｹ蠕∝ｹｳ陦｡
|
||
- 譌<>蠑ょｸｸ逧<EFBDB8><E980A7>扈<EFBFBD>慮髣ｴ讓｡蠑<EFBDA1>
|
||
|
||
### 荳榊ｮ梧紛<E6A2A7><E7B49B>igh/Unclear risk of bias<61>?
|
||
莉･荳区ュ蜀ｵ蛻､螳壻ｸｺ荳榊ｮ梧紛<EFBFBD>?
|
||
- 笶?莉<>署蛻?髫乗惻蛻<E683BB>ｻ<EFBFBD>"菴<>裏蜈ｷ菴捺婿豕<E5A9BF>
|
||
- 笶?菴ｿ逕ｨ荳榊ｽ鍋噪髫乗惻蛹匁婿豕包ｼ域潔譌･譛溘∽ｽ城劼蜿ｷ縲∽ｺ､譖ｿ蛻<EFBDBF><E89BBB><EFBFBD><EFBFBD>
|
||
- 笶?譌<>蛻<EFBFBD><E89BBB>髫占酪謌門<E8AC8C>驟埼嚼阯丈ｸ榊ｽ難ｼ亥ｼ謾ｾ蛻<EFBDBE><E89BBB>陦ｨ<E999A6>?
|
||
- 笶?蝓ｺ郤ｿ蟄伜惠譏ｾ闡嶺ｸ榊ｹｳ陦｡荳疲裏隹<E8A38F><E99AB9>?
|
||
- 笞<><E7AC9E><EFBFBD> 譁ｹ豕墓緒霑ｰ讓｡邉奇ｼ梧裏豕募愛譁ｭ蜈<EFBDAD><E89C88>諤?
|
||
|
||
### 譌<>豕募愛譁ｭ<E8AD81><EFBDAD>nclear risk<73>?
|
||
- 隶ｺ譁<EFBDBA>ｮ悟<EFBDAE>譛ｪ謠仙所髫乗惻蛹匁婿豕<E5A9BF>
|
||
- 莉<>惠蜈ｶ莉門慍譁ｹ<E8AD81>亥ｦよｳｨ蜀梧婿譯茨ｼ画署蛻ｰ<E89BBB>御ｽ<E5BEA1>悽譁<E682BD>悴謠剰ｿｰ
|
||
|
||
## 蜈ｳ髞ｮ菫｡蜿ｷ隸?
|
||
|
||
**鬮倩ｴｨ驥丈ｿ｡蜿ｷ<E89CBF>亥ｮ梧紛<E6A2A7>?*<2A>?
|
||
- "computer-generated random sequence"
|
||
- "central randomization/allocation"
|
||
- "sealed opaque envelopes"
|
||
- "stratified randomization"
|
||
- "block randomization"
|
||
- "minimization"
|
||
- "allocation concealment"
|
||
|
||
**鬟朱勦菫｡蜿ｷ<E89CBF>井ｸ榊ｮ梧紛<E6A2A7>?*<2A>?
|
||
- "alternating allocation"
|
||
- "by date of birth"
|
||
- "by hospital number"
|
||
- "open allocation"
|
||
- "assigned by investigator"
|
||
|
||
## 謠仙叙謖<E58F99>漉
|
||
|
||
1. **莨伜<E88EA8>譟･謇ｾ菴咲ｽｮ**<EFBFBD>?
|
||
- Methods遶<73>闃ら<E99783>?Randomization"蟆剰鰍
|
||
- Figure 1 (CONSORT豬∫ｨ句<EFBDA8>?
|
||
- Trial Registration菫｡諱ｯ
|
||
- 陦･蜈<EFBDA5>攝譁呻ｼ<E591BB>upplementary Materials<6C>?
|
||
|
||
2. **莠､蜿蛾ｪ瑚ｯ<E7919A>**<EFBFBD>?
|
||
- Methods謠剰ｿｰ vs. Results荳ｭ逧<EFBDAD>渕郤ｿ謨ｰ謐ｮ
|
||
- 螢ｰ遘ｰ逧<EFBDB0>婿豕?vs. 螳樣刔逧<E58894>渕郤ｿ蟷ｳ陦｡諠<EFBDA1><E8ABA0>?
|
||
|
||
3. **迚ｹ谿頑ュ蜀ｵ**<EFBFBD>?
|
||
- 螯よ棡謠仙芦"see protocol"謌?see trial registration"<22>梧<EFBFBD><E6A2A7>ｮｰ荳ｺ髴隕∵衍髦<E8A18D>､夜Κ襍<CE9A><E8A58D>?
|
||
- 螯よ棡譏ｯ螟壻ｸｭ蠢<EFBDAD><E8A0A2>皮ｩｶ<EFBDA9>悟ｺ碑ｯ･譛我ｸｭ蠢<EFBDAD>囂譛ｺ蛹也ｳｻ扈?
|
||
|
||
## 霎灘<E99C8E>譬ｼ蠑擾ｼ井ｸ･譬ｼJSON<4F>?
|
||
|
||
{
|
||
"assessment": "螳梧紛" | "荳榊ｮ梧<EFBDAE>? | "譌<>豕募愛譁ｭ",
|
||
"evidence": {
|
||
"quote": "蜴滓枚蠑慕畑<E68595>?00-300蟄暦ｼ悟桁蜷ｫ蜈ｳ髞ｮ譁ｹ豕墓緒霑ｰ<E99C91>?,
|
||
"location": {
|
||
"section": "Methods",
|
||
"subsection": "Randomization",
|
||
"page": 3,
|
||
"paragraph": 2,
|
||
"figure": "Figure 1 (CONSORT diagram)"
|
||
},
|
||
"highlightedKeywords": [
|
||
"蜈ｳ髞ｮ隸?",
|
||
"蜈ｳ髞ｮ隸?"
|
||
]
|
||
},
|
||
"reasoning": "蛻､譁ｭ逅<EFBDAD>罰<EFBFBD>壽<EFBFBD>ｹ謐ｮ蜴滓枚蠑慕畑<E68595>瑚ｯ･遐皮ｩ?..",
|
||
"confidence": 0.95,
|
||
"robAssessment": "Low risk" | "High risk" | "Unclear risk",
|
||
"needsExternalVerification": false,
|
||
"notes": "蜈ｶ莉冶ｯｴ譏趣ｼ亥庄騾会ｼ<E4BC9A>"
|
||
}
|
||
|
||
## 豕ｨ諢丈ｺ矩｡ｹ
|
||
|
||
1. **荳･譬ｼ驕ｵ螳<EFBDB5>ochrane譬<65>㊥**<EFBFBD>壼ｮ∝庄蛻､譁ｭ荳ｺ"荳榊ｮ梧<EFBDAE>?<3F>御ｸ崎ｦ∬ｿ<E288AC>ｺ主ｮｽ譚?
|
||
2. **蠑慕畑蠢<E79591>｡ｻ蜈ｷ菴<EFBDB7>**<EFBFBD>壻ｸ崎ｦ∫ｬｼ扈溷慍隸?Methods遶<73>闃よ署蛻ｰ"<22>悟ｿ<E6829F>｡ｻ扈吝<E68988>蜈ｷ菴灘ｼ慕<EFBDBC>?
|
||
3. **鄂ｮ菫｡蠎ｦ隸壼ｮ?*<2A>壼ｦよ棡菫｡諱ｯ荳肴ｸ<E882B4>匆<EFBFBD>碁剄菴残onfidence蟷ｶ譬<EFBDB6>ｮｰneedsExternalVerification
|
||
4. **蛹ｺ蛻<EFBDBA>"譛ｪ蛛<EFBDAA>"蜥?譛ｪ謚･蜻?**<EFBFBD>?
|
||
- 螯よ棡隶ｺ譁<EFBDBA><E8AD81>遑ｮ隸?no randomization"<22>径ssessment="荳榊ｮ梧<EFBDAE>?
|
||
- 螯よ棡隶ｺ譁<EFBDBA>ｮ悟<EFBDAE>譛ｪ謠仙所<E4BB99>径ssessment="譌<>豕募愛譁ｭ"
|
||
```
|
||
|
||
**蜈ｶ莉<EFBDB6>11荳ｪ蟄玲ｮｵ逧Пrompt讓｡譚ｿ**<EFBFBD>夂ｱｻ莨ｼ扈捺桷<EFBFBD>梧<EFBFBD>ｹ謐ｮCochrane譬<EFBFBD>㊥隹<EFBFBD>紛蛻､譁ｭ譬<EFBFBD>㊥
|
||
|
||
---
|
||
|
||
### 荳峨｀VP謌先悽鬚<E682BD>ｮ<EFBFBD>
|
||
|
||
**蝨ｺ譎ｯ<E8AD8E>?00遽<30><E981BD>譁<EFBFBD>､咲ｭ?*
|
||
|
||
| 邇ｯ闃<EFBDAF> | Token豸郁?| 讓｡蝙<EFBDA1> | 謌先悽 |
|
||
|------|----------|------|------|
|
||
| Nougat謠仙叙 | - | 譛ｬ蝨ｰ讓｡蝙<EFBDA1> | ﾂ･0 |
|
||
| 12蟄玲ｮｵ謠仙叙<E4BB99>亥曙讓｡蝙具ｼ?| 12K ﾃ<> 2 = 24K | DeepSeek-V3 + Qwen3-Max | ﾂ･0.06/遽?|
|
||
| 蜀ｲ遯∝ｭ玲ｮｵ莠ｺ蟾･螟肴<E89E9F>ｸ<EFBFBD>?0%<25>?| - | 莠ｺ蟾･ | 2蛻<32>帖/蟄玲ｮｵ |
|
||
| **100遽<30>ｻ謌先<E8AC8C>?* | - | - | **ﾂ･6 + 莠ｺ蟾･謌先悽** |
|
||
|
||
**蟇ｹ豈<EFBDB9>**<EFBFBD>?
|
||
- 蜈ｨ譁<EFBDA8>ｸ谺｡諤ｧ謠仙叙<E4BB99>堋･10/100遽?
|
||
- 蛻<>ｮｵ謠仙叙<E4BB99>堋?/100遽?
|
||
- **闃ら怐40%謌先悽 + 蜃<>｡ｮ邇<EFBDAE>署蜊?*
|
||
|
||
---
|
||
|
||
### 蝗帙｀VP鬪梧噺譬<E599BA>㊥
|
||
|
||
| 謖<><E8AC96><EFBFBD> | 逶ｮ譬<EFBDAE> | 鬪瑚ｯ∵婿豕<E5A9BF> |
|
||
|------|------|----------|
|
||
| 蟄玲ｮｵ謠仙叙螳梧紛邇?| 竕?90% | 12蟄玲ｮｵ驛ｽ譛臥ｻ捺棡<E68DBA>磯撼"譌<>豕募愛譁ｭ"<22>?|
|
||
| 蜿梧ｨ｡蝙倶ｸ閾ｴ邇<EFBDB4> | 竕?75% | 12蟄玲ｮｵ荳ｭ閾ｳ蟆?荳ｪ荳閾?|
|
||
| 隸∵紺體ｾ螳梧紛諤?| 100% | 豈丈ｸｪ蟄玲ｮｵ驛ｽ譛牙次譁<E6ACA1>ｼ慕畑蜥御ｽ咲ｽ?|
|
||
| 莠ｺ蟾･螟肴<E89E9F>ｸ髦溷<E9ABA6> | 竕?30% | 髴隕∽ｺｺ蟾･莉句<E88E89>逧<EFBFBD>枚迪ｮ蜊<EFBDAE>豈<EFBFBD> |
|
||
| Nougat謌仙粥邇?| 竕?85% | 闍ｱ譁<EFBDB1>ｮｺ譁<EFBDBA><E8AD81>蜉滓署蜿匁ｯ比ｾ<E6AF94> |
|
||
| 螟<>炊騾溷ｺｦ | 竕?3蛻<33>帖/遽?| 莉傘DF蛻ｰ扈捺棡逧<E6A3A1>ｻ譌ｶ髟?|
|
||
|
||
---
|
||
|
||
## <20>嶋 V1.0 髦ｶ谿ｵ<E8B0BF>? 蜻ｨ<E89CBB><EFBDA8>
|
||
|
||
### 逶ｮ譬<EFBDAE>ｮ壻ｽ<E5A3BB>
|
||
|
||
- **蜃<>｡ｮ邇<EFBDAE>岼譬?*<2A>壺翁 92%
|
||
- **菫｡諱ｯ螳梧紛邇?*<2A>壺翁 95%
|
||
- **謌先悽鬚<E682BD>ｮ<EFBFBD>**<2A>壺王 ﾂ･0.08/遽<>ｼ域匱閭ｽ謌先悽莨伜喧<E4BC9C>?
|
||
- **莠､莉俶<E88E89><E4BFB6>㊥**<2A>夐ｫ倩ｴｨ驥剰ｾ灘<EFBDBE><E78198>悟ｮ梧紛隸∵紺體ｾ<E9AB94>梧匱閭ｽ雍ｨ驥乗而蛻?
|
||
|
||
---
|
||
|
||
### 荳縲∬ｴｨ驥乗署蜊<E7BDB2>ｭ也<EFBDAD>?
|
||
|
||
#### 1.1 笨?Cochrane譬<65>㊥Prompt蠅槫ｼｺ
|
||
|
||
**蝨ｨMVP蝓ｺ遑荳雁｢槫<EFBDA2>?*<2A>?
|
||
|
||
1. **Few-shot蛹ｻ蟄ｦ譯井ｾ<E4BA95>**<EFBFBD>域ｯ丈ｸｪ蟄玲ｮ?-5荳ｪ逵溷ｮ樊｡井ｾ具ｼ<E585B7>
|
||
|
||
```markdown
|
||
## 蜿り<E3828A>｡井ｾ?
|
||
|
||
莉･荳区<EFBFBD>?荳ｪ逵溷ｮ朿CT遐皮ｩｶ逧<EFBDB6>囂譛ｺ蛹匁婿豕戊ｯ<E6888A>ｼｰ譯井ｾ具ｼ悟ｸｮ蜉ｩ菴<EFBDA9>逅<EFBFBD>ｧ｣蛻､譁ｭ譬<EFBDAD>㊥<EFBFBD>?
|
||
|
||
### 譯井ｾ<E4BA95>1<EFBFBD>夐ｫ倩ｴｨ驥蹴CT<43><54>EJM, 2023<32>?
|
||
**蜴滓枚蠑慕畑**<2A>?
|
||
"Randomization was performed with the use of a computer-generated sequence
|
||
with stratification according to center and baseline NIHSS score (竕?0 or >10).
|
||
Allocation was concealed through a central web-based system (IWRS)."
|
||
|
||
**隸<>ｼｰ扈捺棡**<EFBFBD>壼ｮ梧<EFBFBD>?
|
||
**逅<>罰**<EFBFBD>?
|
||
1. 笨?譏守｡ｮ逧<EFBDAE>ｺ丞<EFBDBA>逕滓<E98095>譁ｹ豕包ｼ<E58C85>omputer-generated<65>?
|
||
2. 笨?蛻<>ｱる囂譛ｺ蛹厄ｼ域署鬮伜ｹｳ陦｡諤ｧ<E8ABA4><EFBDA7>
|
||
3. 笨?荳ｭ蠢<EFBDAD><E8A0A2>驟埼嚼阯擾ｼ<E693BE>WRS<52>?
|
||
4. 笨?蝓ｺ郤ｿTable 1譏ｾ遉ｺ荳､扈<EFBDA4>ｹｳ陦｡濶ｯ螂ｽ<E89E82><EFBDBD>>0.05<EFBFBD>?
|
||
**RoB 2.0蛻､譁ｭ**<2A>哭ow risk of bias
|
||
|
||
---
|
||
|
||
### 譯井ｾ<E4BA95>2<EFBFBD>夊ｴｨ驥丈ｸ崎ｶｳ<EFBDB6>域汾譛溷<E8AD9B>? 2020<32>?
|
||
**蜴滓枚蠑慕畑**<2A>?
|
||
"Patients were randomly assigned to receive either drug A or placebo
|
||
in a 1:1 ratio. Randomization was performed by the study coordinator."
|
||
|
||
**隸<>ｼｰ扈捺棡**<EFBFBD>壻ｸ榊ｮ梧紛
|
||
**逅<>罰**<EFBFBD>?
|
||
1. 笶?譛ｪ隸ｴ譏主ｺ丞<EFBDBA>逕滓<E98095>譁ｹ豕包ｼ井ｻ<E4BA95>ｯｴ"髫乗惻"<22>?
|
||
2. 笶?逕ｱ遐皮ｩｶ蜊剰ｰ<E589B0>遭謇ｧ陦碁囂譛ｺ蛹厄ｼ域裏蛻<E8A38F><E89BBB>髫占酪<E58DA0><E985AA>
|
||
3. 笞<><E7AC9E><EFBFBD> Table 1譏ｾ遉ｺ蟇ｹ辣ｧ扈<EFBDA7>ｹｴ鮴<EFBDB4>￥螟ｧ<E89E9F><EFBDA7>66.2 vs 62.1, P=0.04<EFBFBD>?
|
||
**RoB 2.0蛻､譁ｭ**<2A>唏igh risk of bias
|
||
**髣ｮ鬚<EFBDAE>**<EFBFBD>壼庄閭ｽ蟄伜惠騾画叫蛛丞?
|
||
|
||
---
|
||
|
||
### 譯井ｾ<E4BA95>3<EFBFBD>夊ｾｹ逡梧ュ蜀ｵ<E89C80><EFBDB5>ancet, 2021<32>?
|
||
**蜴滓枚蠑慕畑**<2A>?
|
||
"Randomization was done with sequentially numbered, opaque, sealed envelopes
|
||
prepared by an independent statistician not otherwise involved in the trial."
|
||
|
||
**隸<>ｼｰ扈捺棡**<EFBFBD>壼ｮ梧<EFBFBD>?
|
||
**逅<>罰**<EFBFBD>?
|
||
1. 笨?陌ｽ髱樔ｸｭ蠢<EFBDAD>囂譛ｺ蛹厄ｼ御ｽ<E5BEA1>ｽｿ逕ｨ蟇<EFBDA8>ｰ∽ｿ｡蟆?
|
||
2. 笨?迢ｬ遶狗ｬｬ荳画婿蜃<E5A9BF>､<EFBFBD>ｼ育ｻ溯ｮ｡蟶茨ｼ<E88CA8>
|
||
3. 笨?荳埼丞<C280><E4B89E><EFBFBD>paque<75>我ｸ泌ｯ<E6B38C>ｰ<EFBFBD>ｼ<EFBFBD>ealed<65>?
|
||
4. 笨?蝓ｺ郤ｿ蟷ｳ陦｡濶ｯ螂ｽ
|
||
**RoB 2.0蛻､譁ｭ**<2A>哭ow risk of bias
|
||
**隸ｴ譏<EFBDB4>**<EFBFBD>夂ｬｦ蜷<EFBFBD>ochrane譬<EFBFBD>㊥<EFBFBD>亥ｯ<EFBFBD>ｰ∽ｿ｡蟆?+ 迢ｬ遶句㊥螟<E38AA5>庄謗･蜿暦ｼ<E69AA6>
|
||
|
||
---
|
||
|
||
邇ｰ蝨ｨ隸ｷ菴<EFBFBD>蜿り<EFBFBD>ｻ･荳頑｡井ｾ狗噪隸<EFBFBD>ｼｰ譁ｹ蠑擾ｼ悟<EFBFBD>譫仙ｽ灘燕隶ｺ譁?..
|
||
```
|
||
|
||
2. **Chain of Thought謗ｨ逅<EFBDA8>**
|
||
|
||
```markdown
|
||
## 霎灘<E99C8E>譬ｼ蠑擾ｼ亥｢槫ｼｺ迚茨ｼ?
|
||
|
||
{
|
||
"assessment": "螳梧紛",
|
||
|
||
// 箝?譁ｰ蠅橸ｼ夐先ｭ･謗ｨ逅<EFBDA8>ｿ<EFBFBD>ｨ<EFBFBD>
|
||
"reasoning_steps": {
|
||
"step1_sequenceGeneration": {
|
||
"finding": "隶ｺ譁<EFBDBA>署蛻ｰ'computer-generated random sequence'",
|
||
"evaluation": "貊｡雜ｳ蠎丞<E8A08E>逕滓<E98095>譁ｹ豕戊ｦ∵ｱ<E288B5> 笨?
|
||
},
|
||
"step2_allocationConcealment": {
|
||
"finding": "菴ｿ逕ｨ'central web-based system (IWRS)'",
|
||
"evaluation": "貊｡雜ｳ蛻<EFBDB3><E89BBB>髫占酪隕∵ｱ<E288B5> 笨?
|
||
},
|
||
"step3_baselineBalance": {
|
||
"finding": "Table 1譏ｾ遉ｺ荳ｻ隕∫音蠕￣>0.05",
|
||
"evaluation": "譌<>譏取仞騾画叫蛛丞夊ｯ∵<EFBDAF>?笨?
|
||
},
|
||
"step4_finalJudgment": {
|
||
"conclusion": "荳蛾｡ｹ譬<EFBDB9>㊥蝮<E38AA5>ｻ｡雜ｳ<E99B9C>悟愛譁ｭ荳?螳梧紛'",
|
||
"confidence": 0.95
|
||
}
|
||
},
|
||
|
||
"evidence": { ... },
|
||
"robAssessment": "Low risk"
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
#### 1.2 笨?蜈ｨ譁<EFBDA8>ｺ､蜿蛾ｪ瑚ｯ<E7919A>ｼ磯亟驕玲ｼ擾ｼ?
|
||
|
||
**蝨ｨ蛻<EFBDA8>ｮｵ謠仙叙蜷趣ｼ悟｢槫刈蜈ｨ譁<EFBDA8>ｪ瑚ｯ∫識闃?*<2A>?
|
||
|
||
```typescript
|
||
// 髦ｶ谿ｵ1<EFBDB5>壼<EFBFBD>谿ｵ謠仙叙<E4BB99>亥ｷｲ螳梧<E89EB3><E6A2A7><EFBFBD>
|
||
const segmentedResults = await extractAllFieldsSegmented(sections);
|
||
|
||
// 箝?髦ｶ谿ｵ2<EFBDB5>壼<EFBFBD>譁<EFBFBD>ｺ､蜿蛾ｪ瑚ｯ<E7919A>ｼ域眠蠅橸ｼ?
|
||
async function crossValidateWithFullText(
|
||
segmentedResults: FieldResult[],
|
||
fullTextMarkdown: string
|
||
): Promise<ValidationReport> {
|
||
|
||
// 鬪瑚ｯ<E7919A>1<EFBFBD>壽｣譟･譏ｯ蜷ｦ譛蛾＠貍丈ｿ｡諱ｯ
|
||
const missingInfoChecks = await Promise.all([
|
||
checkForMissingInfo('髫乗惻蛹匁婿豕?, fullTextMarkdown, segmentedResults),
|
||
checkForMissingInfo('逶ｲ豕<EFBFBD>', fullTextMarkdown, segmentedResults),
|
||
// ... 蜈ｶ莉門<E88E89>髞ｮ蟄玲ｮｵ
|
||
]);
|
||
|
||
// 鬪瑚ｯ<E7919A>2<EFBFBD>壽｣譟･譏ｯ蜷ｦ譛臥泝逶ｾ菫｡諱ｯ
|
||
const contradictionChecks = await checkContradictions(
|
||
segmentedResults,
|
||
fullTextMarkdown
|
||
);
|
||
|
||
// 鬪瑚ｯ<E7919A>3<EFBFBD>壽｣譟･譏ｯ蜷ｦ謠仙芦陦･蜈<EFBDA5>攝譁?
|
||
const supplementaryCheck = checkSupplementaryMaterial(fullTextMarkdown);
|
||
|
||
return {
|
||
missingInfoAlerts: missingInfoChecks.filter(c => c.hasIssue),
|
||
contradictions: contradictionChecks,
|
||
needsSupplementary: supplementaryCheck.needsExternal,
|
||
overallCompleteness: calculateCompleteness(...)
|
||
};
|
||
}
|
||
|
||
// 遉ｺ萓具ｼ壽｣譟･驕玲ｼ丈ｿ｡諱?
|
||
async function checkForMissingInfo(
|
||
field: string,
|
||
fullText: string,
|
||
extractedResult: FieldResult
|
||
): Promise<ValidationAlert> {
|
||
|
||
// 螯よ棡蟾ｲ扈丞愛螳壻ｸ?螳梧紛"<22>瑚ｷｳ霑?
|
||
if (extractedResult.assessment === '螳梧紛') {
|
||
return { field, hasIssue: false };
|
||
}
|
||
|
||
// 蝨ｨ蜈ｨ譁<EFBDA8>ｸｭ謳懃ｴ｢蜈ｳ髞ｮ隸?
|
||
const keywords = FIELD_KEYWORDS[field]; // 鬚<>ｮ壻ｹ牙<EFBDB9>髞ｮ隸崎｡?
|
||
const foundKeywords = keywords.filter(kw =>
|
||
fullText.toLowerCase().includes(kw.toLowerCase())
|
||
);
|
||
|
||
// 螯よ棡蜈ｨ譁<EFBDA8>ｸｭ譛牙<E8AD9B>髞ｮ隸搾ｼ御ｽ<E5BEA1>署蜿也ｻ捺棡譏ｯ"譌<>豕募愛譁ｭ"
|
||
if (foundKeywords.length > 0 && extractedResult.assessment === '譌<EFBFBD>豕募愛譁ｭ') {
|
||
return {
|
||
field,
|
||
hasIssue: true,
|
||
severity: 'warning',
|
||
message: `蜈ｨ譁<EFBDA8>ｸｭ蜿醍鴫蜈ｳ髞ｮ隸阪?{foundKeywords.join(', ')}縲托ｼ<E68998>
|
||
菴<>ｭ玲ｮ?${field}"蛻､譁ｭ荳?譌<>豕募愛譁ｭ"<22>悟庄閭ｽ蟄伜惠驕玲ｼ汁,
|
||
suggestedAction: 'targeted_re_extraction',
|
||
keywords: foundKeywords
|
||
};
|
||
}
|
||
|
||
return { field, hasIssue: false };
|
||
}
|
||
```
|
||
|
||
**謨域棡**<EFBFBD>?
|
||
- 驕玲ｼ丈ｿ｡諱ｯ譽蜃ｺ邇<EFBDBA>ｼ?% 竊?80%
|
||
- 蜃<>｡ｮ邇<EFBDAE>署蜊<E7BDB2>ｼ<EFBFBD>85% 竊?92%
|
||
|
||
---
|
||
|
||
#### 1.3 笨?蛹ｻ蟄ｦ騾ｻ霎題ｧ<E9A18C><EFBDA7>蠑墓梼
|
||
|
||
**閾ｪ蜉ｨ譽譟･蟶ｸ隗∫噪騾ｻ霎鷹漠隸ｯ**<EFBFBD>?
|
||
|
||
```typescript
|
||
const MEDICAL_LOGIC_RULES = [
|
||
{
|
||
id: 'rule_001',
|
||
name: 'RCT蠢<54>｡ｻ譛蛾囂譛ｺ蛹<EFBDBA>',
|
||
check: (data) => {
|
||
const isRCT = data.遐皮ｩｶ隶ｾ隶｡.toLowerCase().includes('rct') ||
|
||
data.遐皮ｩｶ隶ｾ隶｡.includes('髫乗惻');
|
||
const hasRandomization = data.髫乗惻蛹匁婿豕?!== '譌<>豕募愛譁ｭ';
|
||
return !isRCT || hasRandomization;
|
||
},
|
||
severity: 'error',
|
||
message: '遐皮ｩｶ螢ｰ遘ｰ譏ｯRCT菴<54>悴謇ｾ蛻ｰ髫乗惻蛹匁婿豕墓緒霑?,
|
||
action: 'flag_for_urgent_review'
|
||
},
|
||
|
||
{
|
||
id: 'rule_002',
|
||
name: '蜿檎峇遐皮ｩｶ蠢<EFBFBD>｡ｻ隸ｴ譏守峇豕<EFBFBD>',
|
||
check: (data) => {
|
||
const isDoubleBlind = data.遐皮ｩｶ隶ｾ隶｡.includes('蜿檎峇') ||
|
||
data.遐皮ｩｶ隶ｾ隶｡.includes('double-blind');
|
||
const hasBlinding = data.逶ｲ豕<EFBDB2> !== '譌<EFBFBD>豕募愛譁ｭ' &&
|
||
data.逶ｲ豕<EFBDB2> !== '荳榊ｮ梧<EFBFBD>?;
|
||
return !isDoubleBlind || hasBlinding;
|
||
},
|
||
severity: 'error',
|
||
message: '螢ｰ遘ｰ蜿檎峇菴<E5B387>峇豕墓緒霑ｰ荳榊ｮ梧紛',
|
||
action: 'flag_for_review'
|
||
},
|
||
|
||
{
|
||
id: 'rule_003',
|
||
name: '譬ｷ譛ｬ驥丈ｸ主渕郤ｿ謨ｰ謐ｮ荳閾ｴ諤?,
|
||
check: (data) => {
|
||
const planned = extractNumber(data.譬ｷ譛ｬ驥剰ｮ｡邂?;
|
||
const enrolled = extractNumber(data.遐皮ｩｶ莠ｺ鄒､);
|
||
if (!planned || !enrolled) return true; // 譌<>豕墓署蜿門<E89CBF>霍ｳ霑?
|
||
|
||
const deviation = Math.abs(planned - enrolled) / planned;
|
||
return deviation < 0.3; // 蛛丞ｷｮ<30%
|
||
},
|
||
severity: 'warning',
|
||
message: '隶｡蛻呈<EFBFBD>ｷ譛ｬ驥丈ｸ主ｮ樣刔蜈･扈<EFBFBD>ｷｮ蠑りｾ<EFBFBD>､ｧ(>30%)',
|
||
action: 'add_note'
|
||
},
|
||
|
||
{
|
||
id: 'rule_004',
|
||
name: '蝓ｺ郤ｿ荳榊ｹｳ陦｡髴隕∬ｰ<EFBFBD><EFBFBD>?,
|
||
check: (data) => {
|
||
const hasImbalance = data.蝓ｺ郤ｿ蜿ｯ豈疲?includes('荳榊ｹｳ陦?) ||
|
||
data.蝓ｺ郤ｿ蜿ｯ豈疲?includes('P<0.05');
|
||
const hasAdjustment = data.扈灘ｱ謖<C280><E8AC96><EFBFBD>.includes('隹<EFBFBD>紛') ||
|
||
data.扈灘ｱ謖<C280><E8AC96><EFBFBD>.includes('adjusted');
|
||
return !hasImbalance || hasAdjustment;
|
||
},
|
||
severity: 'warning',
|
||
message: '蝓ｺ郤ｿ蟄伜惠荳榊ｹｳ陦｡菴<EFBFBD>悴隗∬ｰ<EFBFBD>紛蛻<EFBFBD>梵',
|
||
action: 'add_note'
|
||
},
|
||
|
||
{
|
||
id: 'rule_005',
|
||
name: 'ITT蛻<EFBFBD>梵螳梧紛諤?,
|
||
check: (data) => {
|
||
const hasDropout = extractNumber(data.扈捺棡螳梧紛諤? > 0;
|
||
const hasITT = data.扈捺棡螳梧紛諤?toLowerCase().includes('itt') ||
|
||
data.扈捺棡螳梧紛諤?includes('intention-to-treat');
|
||
return !hasDropout || hasITT;
|
||
},
|
||
severity: 'warning',
|
||
message: '蟄伜惠螟ｱ隶ｿ菴<EFBDBF>悴譏守｡ｮITT蛻<54>梵',
|
||
action: 'flag_for_review'
|
||
}
|
||
];
|
||
|
||
// 閾ｪ蜉ｨ鬪瑚ｯ<E7919A>
|
||
function validateMedicalLogic(extractedData: ExtractionResult): LogicReport {
|
||
const violations = [];
|
||
|
||
for (const rule of MEDICAL_LOGIC_RULES) {
|
||
try {
|
||
const passed = rule.check(extractedData);
|
||
if (!passed) {
|
||
violations.push({
|
||
ruleId: rule.id,
|
||
ruleName: rule.name,
|
||
severity: rule.severity,
|
||
message: rule.message,
|
||
action: rule.action
|
||
});
|
||
}
|
||
} catch (error) {
|
||
console.error(`隗<EFBFBD><EFBFBD>${rule.id}謇ｧ陦悟､ｱ雍･:`, error);
|
||
}
|
||
}
|
||
|
||
return {
|
||
totalRules: MEDICAL_LOGIC_RULES.length,
|
||
passedRules: MEDICAL_LOGIC_RULES.length - violations.length,
|
||
violations,
|
||
overallValidity: violations.filter(v => v.severity === 'error').length === 0
|
||
};
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
#### 1.4 笨?螳梧紛隸∵紺體ｾ<E9AB94>亥｢槫ｼｺ迚茨ｼ<E88CA8>
|
||
|
||
**V1.0隕∵ｱ<EFBFBD>**<EFBFBD>壻ｸ堺ｻ<EFBFBD>怏蠑慕畑<EFBFBD>瑚ｿ倩ｦ∵怏蜈ｷ菴灘ｮ壻ｽ榊柱鬮倅ｺ?
|
||
|
||
```typescript
|
||
interface EnhancedEvidence {
|
||
field: string;
|
||
assessment: string;
|
||
|
||
evidence: {
|
||
// 荳ｻ隕∬ｯ∵紺
|
||
primaryQuote: {
|
||
text: string; // 蜴滓枚蠑慕畑
|
||
location: {
|
||
section: string; // "Methods"
|
||
subsection?: string; // "Randomization"
|
||
page: number; // 3
|
||
paragraph: number; // 2
|
||
lineRange?: [number, number]; // [45, 52]
|
||
};
|
||
highlightedText: string; // HTML鬮倅ｺｮ迚域悽
|
||
keywords: string[]; // 蜈ｳ髞ｮ隸榊<E99AB8>陦?
|
||
};
|
||
|
||
// 謾ｯ謖∬ｯ∵紺<E288B5>亥庄騾会ｼ<E4BC9A>
|
||
supportingQuotes?: Array<{
|
||
text: string;
|
||
location: any;
|
||
relation: string; // "confirms" | "contradicts" | "complements"
|
||
}>;
|
||
|
||
// 陦ｨ譬ｼ/蝗ｾ迚<EFBDBE>ｯ∵紺
|
||
tableEvidence?: {
|
||
tableName: string; // "Table 1"
|
||
relevantCells: string[]; // 逶ｸ蜈ｳ蜊募<E89C8A>譬ｼ蜀<EFBDBC>ｮ?
|
||
interpretation: string; // 蟇ｹ陦ｨ譬ｼ逧<EFBDBC>ｧ｣隸ｻ
|
||
};
|
||
|
||
figureEvidence?: {
|
||
figureName: string; // "Figure 1"
|
||
caption: string;
|
||
relevantInfo: string;
|
||
};
|
||
};
|
||
|
||
// 箝?譁ｰ蠅橸ｼ壼ｮ梧紛謗ｨ逅<EFBDA8>得
|
||
reasoningChain: {
|
||
cochraneCriteria: string[]; // 蠎皮畑逧Гochrane譬<65>㊥
|
||
keyFindings: string[]; // 蜈ｳ髞ｮ蜿醍鴫
|
||
assessment: string; // 譛扈亥愛譁?
|
||
confidence: number;
|
||
uncertainties?: string[]; // 荳咲｡ｮ螳壼屏邏?
|
||
};
|
||
|
||
// 箝?譁ｰ蠅橸ｼ壼庄霑ｽ貅ｯ諤ｧ蜈<EFBDA7>焚謐ｮ
|
||
metadata: {
|
||
extractionTimestamp: string;
|
||
modelUsed: string;
|
||
promptVersion: string;
|
||
processingTime: number;
|
||
};
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
### 莠後〃1.0謌先悽鬚<EFBFBD>ｮ<EFBFBD>
|
||
|
||
**蝨ｺ譎ｯ<E8AD8E>?00遽<30><E981BD>譁<EFBFBD>､咲ｭ?*
|
||
|
||
| 邇ｯ闃<EFBDAF> | Token豸郁?| 讓｡蝙<EFBDA1> | 謌先悽 |
|
||
|------|----------|------|------|
|
||
| 12蟄玲ｮｵ蛻<EFBDB5>ｮｵ謠仙叙<E4BB99>亥曙讓｡蝙具ｼ?| 12K | DeepSeek-V3 + Qwen3-Max | ﾂ･0.06/遽?|
|
||
| 蜈ｨ譁<EFBDA8>ｺ､蜿蛾ｪ瑚ｯ<E7919A> | 3K | DeepSeek-V3 | ﾂ･0.003/遽?|
|
||
| 蜈ｳ髞ｮ蟄玲ｮｵ陦･蜈<EFBDA5>署蜿厄ｼ?0%<25>?| 2K | Qwen3-Max | ﾂ･0.016/遽<>ｼ井ｻ?0%譁<>鍵<EFBFBD>?|
|
||
| **100遽<30>ｻ謌先<E8AC8C>?* | - | - | **ﾂ･7.9** |
|
||
|
||
**雍ｨ驥乗署蜊<E7BDB2>**<EFBFBD>壼㊥遑ｮ邇<EFBFBD> 85% 竊?92%
|
||
**謌先悽蠅槫刈**<EFBFBD>堋? 竊?ﾂ･8<EFBDA5>?33%<25>御ｽ<E5BEA1>ｴｨ驥乗仞闡玲署蜊<E7BDB2>ｼ?
|
||
|
||
---
|
||
|
||
### 荳峨〃1.0鬪梧噺譬<EFBFBD>㊥
|
||
|
||
| 謖<><E8AC96><EFBFBD> | 逶ｮ譬<EFBDAE> | 鬪瑚ｯ∵婿豕<E5A9BF> |
|
||
|------|------|----------|
|
||
| 蜃<>｡ｮ邇<EFBDAE>ｼ井ｺｺ蟾･謚ｽ譟･<E8AD9F>?| 竕?92% | 髫乗惻謚ｽ譟･50遽<30>ｼ御ｸ灘ｮｶ隸<EFBDB6>ｼｰ |
|
||
| 菫｡諱ｯ螳梧紛邇?| 竕?95% | 12蟄玲ｮｵ蝮<EFBDB5>怏譛画譜扈捺棡 |
|
||
| 隸∵紺體ｾ螳梧紛諤?| 100% | 豈丈ｸｪ蟄玲ｮｵ譛芽ｯｦ扈<EFBDA6>ｯ∵紺蜥梧耳逅<E880B3><E98085>?|
|
||
| 驕玲ｼ丈ｿ｡諱ｯ譽蜃ｺ邇<EFBDBA> | 竕?80% | 莠､蜿蛾ｪ瑚ｯ∝書邇ｰ逧<EFBDB0>＠貍乗ｯ比ｾ?|
|
||
| 騾ｻ霎題ｧ<E9A18C><EFBDA7>隕<EFBFBD>尠邇?| 竕?80% | 隗<><E99A97>蠑墓梼譽譟･騾夊ｿ<E5A48A><EFBDBF>?|
|
||
| 莠ｺ蟾･螟肴<E89E9F>ｸ髦溷<E9ABA6> | 竕?25% | 髴隕∽ｺｺ蟾･莉句<E88E89>逧<EFBFBD>枚迪ｮ蜊<EFBDAE>豈<EFBFBD> |
|
||
|
||
---
|
||
|
||
## <20>醇 V2.0 髦ｶ谿ｵ<E8B0BF>? 蜻ｨ<E89CBB><EFBDA8>
|
||
|
||
### 逶ｮ譬<EFBDAE>ｮ壻ｽ<E5A3BB>
|
||
|
||
- **蜃<>｡ｮ邇<EFBDAE>岼譬?*<2A>壺翁 96%<25>亥現蟄ｦ郤ｧ譬<EFBDA7>㊥<EFBFBD>?
|
||
- **莠ｺ譛ｺ荳閾ｴ諤?*<2A>咾ohen's Kappa 竕?0.90
|
||
- **謌先悽鬚<E682BD>ｮ<EFBFBD>**<2A>壽潔髴驟咲ｽｮ<EFBDBD>郁ｴｨ驥丈ｼ伜<EFBDBC><E4BC9C><EFBFBD>
|
||
- **莠､莉俶<E88E89><E4BFB6>㊥**<2A>夊<EFBFBD>蜉ｨ蛹冶ｴｨ驥丞ｮ｡隶｡<E99AB6>檎ｬｦ蜷<EFBDA6>ochrane蜿題｡ｨ譬<EFBDA8>㊥
|
||
|
||
---
|
||
|
||
### 荳縲∝現蟄ｦ郤ｧ雍ｨ驥丈ｿ晞囿
|
||
|
||
#### 1.1 笨?荳画ｨ｡蝙倶ｻｲ陬∵惻蛻?
|
||
|
||
**蜈ｳ髞ｮ蟄玲ｮｵ蜀ｲ遯∵慮<E288B5>悟星逕ｨ隨ｬ荳画婿莉ｲ陬?*<2A>?
|
||
|
||
```typescript
|
||
async function threeModelArbitration(
|
||
conflict: FieldConflict,
|
||
relevantContent: string
|
||
) {
|
||
|
||
// 隨ｬ荳画婿莉ｲ陬<EFBDB2>ｼ咾laude-4.5<EFBFBD>磯ｫ倩ｴｨ驥乗ｨ｡蝙具ｼ?
|
||
const arbitrationPrompt = `
|
||
菴<EFBFBD>譏ｯCochrane邉ｻ扈溯ｯ<EFBFBD>ｻｷ荳灘ｮｶ<EFBFBD>檎鴫譛我ｸ､荳ｪAI讓｡蝙句ｯｹ蜷御ｸ蟄玲ｮｵ逧<EFBFBD>愛譁ｭ蟄伜惠蜀ｲ遯<EFBFBD>ｼ<EFBFBD>
|
||
隸ｷ菴<EFBFBD>莉主ｾｪ隸∝現蟄ｦ逧<EFBFBD>ｧ貞ｺｦ扈吝<EFBFBD>譚<EFBFBD>ｨ∝愛譁ｭ縲?
|
||
|
||
縲仙<EFBFBD>遯∝ｭ玲ｮｵ縲托ｼ<EFBFBD>${conflict.field}
|
||
|
||
縲先ｨ｡蝙帰蛻､譁ｭ縲托ｼ<EFBFBD>${conflict.modelA.assessment}
|
||
隸∵紺<EFBFBD>?{conflict.modelA.evidence.quote}
|
||
逅<EFBFBD>罰<EFBFBD>?{conflict.modelA.reasoning}
|
||
鄂ｮ菫｡蠎ｦ<EFBFBD><EFBFBD>${conflict.modelA.confidence}
|
||
|
||
縲先ｨ｡蝙毅蛻､譁ｭ縲托ｼ<EFBFBD>${conflict.modelB.assessment}
|
||
隸∵紺<EFBFBD>?{conflict.modelB.evidence.quote}
|
||
逅<EFBFBD>罰<EFBFBD>?{conflict.modelB.reasoning}
|
||
鄂ｮ菫｡蠎ｦ<EFBFBD><EFBFBD>${conflict.modelB.confidence}
|
||
|
||
縲仙次譁<EFBFBD>托ｼ<EFBFBD>
|
||
${relevantContent}
|
||
|
||
縲蝉ｻｲ陬∽ｻｻ蜉｡縲托ｼ<EFBFBD>
|
||
1. 譬ｹ謐ｮCochrane RoB 2.0譬<EFBFBD>㊥<EFBFBD>檎ｻ吝<EFBFBD>菴<EFBFBD>逧<EFBFBD>愛譁?
|
||
2. 蛻<>梵荳､荳ｪ讓｡蝙狗噪蛻､譁ｭ<E8AD81>梧欠蜃ｺ蜩ｪ荳ｪ譖ｴ蜃<EFBDB4>｡ｮ<EFBDA1>域<EFBFBD>驛ｽ荳榊㊥遑ｮ<E98191><EFBDAE>
|
||
3. 蠑慕畑Cochrane謇句<E8AC87>逶ｸ蜈ｳ譚｡谺ｾ謾ｯ謖∽ｽ<E288BD>逧<EFBFBD>愛譁ｭ
|
||
4. 螯よ棡莉堺ｸ咲｡ｮ螳夲ｼ梧<EFBDBC>遑ｮ謖<EFBDAE><E8AC96>髴隕∽ｺｺ蟾･螟肴<E89E9F>ｸ逧<EFBDB8>次蝗<E6ACA1>
|
||
|
||
縲占ｾ灘<EFBFBD>譬ｼ蠑上托ｼ哽SON
|
||
`;
|
||
|
||
const arbitrationResult = await llmService.chat(
|
||
'claude-4.5',
|
||
arbitrationPrompt
|
||
);
|
||
|
||
return {
|
||
field: conflict.field,
|
||
arbitrator: 'claude-4.5',
|
||
finalJudgment: arbitrationResult.assessment,
|
||
analysis: {
|
||
modelAAccuracy: arbitrationResult.modelA_correct,
|
||
modelBAccuracy: arbitrationResult.modelB_correct,
|
||
correctModel: arbitrationResult.agree_with,
|
||
cochraneCitation: arbitrationResult.cochrane_reference
|
||
},
|
||
confidence: arbitrationResult.confidence,
|
||
stillNeedsHumanReview: arbitrationResult.confidence < 0.9
|
||
};
|
||
}
|
||
```
|
||
|
||
**謌先悽謗ｧ蛻ｶ**<EFBFBD>?
|
||
- 莉<>惠蜈ｳ髞ｮ蟄玲ｮｵ蜀ｲ遯∵慮蜷ｯ逕ｨ<E98095>磯｢<E7A3AF>ｮ｡10-15%<25>?
|
||
- 蜊墓ｬ｡莉ｲ陬∵<E999AC>譛ｬ<E8AD9B>堋?.02<EFBFBD><EFBFBD>laude-4.5<EFBFBD>?
|
||
- 100遽<30>ｻ鬚晏､匁<EFBDA4>譛ｬ<E8AD9B>堋･2-3
|
||
|
||
---
|
||
|
||
#### 1.2 笨?HITL譎ｺ閭ｽ蛻<EFBDBD>ｵ<EFBFBD>
|
||
|
||
**蝓ｺ莠手ｧ<E6898B><EFBDA7>逧<EFBFBD>匱閭ｽ莨伜<E88EA8>郤ｧ謗貞ｺ<E8B29E>**<EFBFBD>?
|
||
|
||
```typescript
|
||
function intelligentTriage(
|
||
extractionResult: ExtractionResult,
|
||
validationReport: ValidationReport,
|
||
arbitrationResults?: ArbitrationResult[]
|
||
): TriageDecision {
|
||
|
||
let priority = 0;
|
||
let needReview = false;
|
||
const reasons = [];
|
||
|
||
// 隗<><E99A97>1<EFBFBD>壻ｸ画ｨ｡蝙倶ｻ堺ｸ堺ｸ閾?竊?譛鬮倅ｼ伜<EFBDBC>郤ｧ
|
||
if (arbitrationResults?.some(a => a.stillNeedsHumanReview)) {
|
||
priority = 100;
|
||
needReview = true;
|
||
reasons.push('荳画ｨ｡蝙倶ｻｲ陬∝錘莉榊ｭ伜惠荳咲｡ｮ螳壽?);
|
||
}
|
||
|
||
// 隗<><E99A97>2<EFBFBD>壼<EFBFBD>髞ｮ蟄玲ｮｵ雍ｨ驥城琉鬚?竊?鬮倅ｼ伜<EFBDBC>郤ｧ
|
||
const criticalIssues = validationReport.violations.filter(v =>
|
||
v.severity === 'error' &&
|
||
FIELD_IMPORTANCE.critical.includes(v.field)
|
||
);
|
||
if (criticalIssues.length > 0) {
|
||
priority = Math.max(priority, 90);
|
||
needReview = true;
|
||
reasons.push(`蜈ｳ髞ｮ蟄玲ｮｵ蟄伜惠雍ｨ驥城琉鬚<E79089>: ${criticalIssues.map(i => i.field).join(', ')}`);
|
||
}
|
||
|
||
// 隗<><E99A97>3<EFBFBD>啌CT遐皮ｩｶ 竊?荳ｭ遲我ｼ伜<EFBDBC>郤ｧ<E983A4>郁ｴｨ驥剰ｦ∵ｱるｫ假ｼ<E58187>
|
||
if (extractionResult.遐皮ｩｶ隶ｾ隶｡.includes('RCT')) {
|
||
priority = Math.max(priority, 70);
|
||
// RCT螯よ棡鄂ｮ菫｡蠎ｦ菴取燕髴隕∝､肴<EFBDA4>?
|
||
if (extractionResult.overallConfidence < 0.9) {
|
||
needReview = true;
|
||
reasons.push('RCT遐皮ｩｶ菴<EFBFBD>紛菴鍋ｽｮ菫｡蠎ｦ菴惹ｺ<EFBFBD>0.9');
|
||
}
|
||
}
|
||
|
||
// 隗<><E99A97>4<EFBFBD>壼<EFBFBD>髞ｮ扈灘ｱ謖<C280><E8AC96><EFBFBD>ｼ域ｭｻ莠｡邇<EFBDA1>ｼ俄<EFBDBC> 鬮倅ｼ伜<EFBDBC>郤ｧ
|
||
if (extractionResult.扈灘ｱ謖<C280><E8AC96><EFBFBD>.includes('豁ｻ莠｡') ||
|
||
extractionResult.扈灘ｱ謖<C280><E8AC96><EFBFBD>.includes('mortality')) {
|
||
priority = Math.max(priority, 80);
|
||
if (extractionResult.扈捺棡螳梧紛諤?!== '螳梧紛') {
|
||
needReview = true;
|
||
reasons.push('蜈ｳ髞ｮ扈灘ｱ謖<EFBFBD><EFBFBD><EFBFBD>ｼ域ｭｻ莠｡邇<EFBFBD>ｼ我ｽ<EFBFBD>ｻ捺棡螳梧紛諤ｧ譛蛾琉鬚<EFBFBD>');
|
||
}
|
||
}
|
||
|
||
// 隗<><E99A97>5<EFBFBD>夐ｫ倡ｽｮ菫｡蠎?+ 譌<>蜀ｲ遯?竊?閾ｪ蜉ｨ騾夊ｿ<E5A48A>
|
||
if (extractionResult.overallConfidence > 0.95 &&
|
||
validationReport.violations.length === 0 &&
|
||
!arbitrationResults) {
|
||
priority = 10;
|
||
needReview = false;
|
||
reasons.push('鬮倩ｴｨ驥乗署蜿厄ｼ梧裏髴莠ｺ蟾･螟肴<EFBFBD>ｸ');
|
||
}
|
||
|
||
// 隗<><E99A97>6<EFBFBD>壼書陦ｨ蝨ｨ鬘ｶ郤ｧ譛溷<E8AD9B> 竊?髯堺ｽ主､肴<EFBDA4>ｸ莨伜<E88EA8>郤?
|
||
const topJournals = ['NEJM', 'Lancet', 'JAMA', 'BMJ'];
|
||
if (topJournals.some(j => extractionResult.metadata.journal?.includes(j))) {
|
||
priority = Math.max(0, priority - 20);
|
||
reasons.push('蜿題｡ｨ蝨ｨ鬘ｶ郤ｧ譛溷<EFBFBD><EFBFBD>梧婿豕募ｭｦ雍ｨ驥城壼ｸｸ霎<EFBFBD>ｫ<EFBFBD>');
|
||
}
|
||
|
||
return {
|
||
priority,
|
||
needReview,
|
||
reasons,
|
||
estimatedReviewTime: estimateReviewTime(extractionResult, needReview),
|
||
reviewDeadline: calculateDeadline(priority)
|
||
};
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
#### 1.3 笨?閾ｪ蜉ｨ雍ｨ驥丞ｮ｡隶｡
|
||
|
||
**螳壽悄謇ｹ驥乗歓譟･<E8AD9F>?0%<25>会ｼ瑚<EFBDBC>蜉ｨ逕滓<E98095>雍ｨ驥乗冠蜻<E586A0>**<EFBFBD>?
|
||
|
||
```typescript
|
||
// 豈丞捉閾ｪ蜉ｨ螳｡隶｡
|
||
async function weeklyQualityAudit(
|
||
startDate: Date,
|
||
endDate: Date
|
||
): Promise<QualityAuditReport> {
|
||
|
||
// 1. 闔ｷ蜿匁悽蜻ｨ謇譛画署蜿也ｻ捺<EFBDBB>?
|
||
const weeklyExtractions = await db.fulltextScreeningResults.findMany({
|
||
where: {
|
||
createdAt: { gte: startDate, lte: endDate }
|
||
}
|
||
});
|
||
|
||
// 2. 髫乗惻謚ｽ譬ｷ10%
|
||
const sampleSize = Math.ceil(weeklyExtractions.length * 0.1);
|
||
const sample = randomSample(weeklyExtractions, sampleSize);
|
||
|
||
// 3. 莠ｺ蟾･螟肴<E89E9F>ｸ譬ｷ譛ｬ
|
||
const humanReviews = await requestHumanReview(sample);
|
||
|
||
// 4. 隶｡邂苓ｴｨ驥乗欠譬<E6ACA0>
|
||
const metrics = {
|
||
蜃<EFBFBD>｡ｮ邇? calculateAccuracy(sample, humanReviews),
|
||
莠ｺ譛ｺ荳閾ｴ諤? calculateCohenKappa(sample, humanReviews),
|
||
蛛<EFBFBD>亠諤ｧ邇<EFBFBD>: calculateFalsePositiveRate(sample, humanReviews),
|
||
蛛<EFBFBD>亢諤ｧ邇<EFBFBD>: calculateFalseNegativeRate(sample, humanReviews),
|
||
|
||
// 蛻<>ｭ玲ｮｵ蜃<EFBDB5>｡ｮ邇<EFBDAE>
|
||
蟄玲ｮｵ蜃<EFBFBD>｡ｮ邇? FIELD_LIST.map(field => ({
|
||
field,
|
||
accuracy: calculateFieldAccuracy(field, sample, humanReviews)
|
||
}))
|
||
};
|
||
|
||
// 5. 讓｡蝙区ｧ閭ｽ蟇ｹ豈<EFBDB9>
|
||
const modelPerformance = {
|
||
'deepseek-v3': analyzeModelPerformance('deepseek-v3', sample, humanReviews),
|
||
'qwen-max': analyzeModelPerformance('qwen-max', sample, humanReviews),
|
||
'claude-4.5': analyzeModelPerformance('claude-4.5', sample, humanReviews)
|
||
};
|
||
|
||
// 6. 髣ｮ鬚伜<E9AC9A>譫<EFBFBD>
|
||
const issues = identifyCommonIssues(sample, humanReviews);
|
||
|
||
// 7. 謾ｹ霑帛ｻｺ隶ｮ
|
||
const recommendations = generateRecommendations(metrics, issues);
|
||
|
||
return {
|
||
period: { start: startDate, end: endDate },
|
||
totalExtractions: weeklyExtractions.length,
|
||
sampledExtractions: sampleSize,
|
||
metrics,
|
||
modelPerformance,
|
||
issues,
|
||
recommendations,
|
||
generatedAt: new Date()
|
||
};
|
||
}
|
||
|
||
// 閾ｪ蜉ｨ隸<EFBDA8>悪蟶ｸ隗<EFBDB8>琉鬚<E79089>
|
||
function identifyCommonIssues(
|
||
sample: Extraction[],
|
||
humanReviews: HumanReview[]
|
||
): Issue[] {
|
||
|
||
const issues = [];
|
||
|
||
// 髣ｮ鬚<EFBDAE>1<EFBFBD>壽汾荳ｪ蟄玲ｮｵ髞呵ｯｯ邇<EFBDAF>ｫ?
|
||
for (const field of FIELD_LIST) {
|
||
const fieldErrors = countFieldErrors(field, sample, humanReviews);
|
||
if (fieldErrors / sample.length > 0.15) { // 髞呵ｯｯ邇?15%
|
||
issues.push({
|
||
type: 'high_field_error_rate',
|
||
field,
|
||
errorRate: fieldErrors / sample.length,
|
||
examples: getErrorExamples(field, sample, humanReviews, 3),
|
||
recommendation: `莨伜喧蟄玲ｮｵ"${field}"逧Пrompt讓｡譚ｿ謌芳ew-shot譯井ｾ義
|
||
});
|
||
}
|
||
}
|
||
|
||
// 髣ｮ鬚<EFBDAE>2<EFBFBD>夂音螳夂ｱｻ蝙狗<E89D99>皮ｩｶ髞呵ｯｯ邇<EFBDAF>ｫ?
|
||
const studyTypeErrors = analyzeByStudyType(sample, humanReviews);
|
||
for (const [studyType, errorRate] of Object.entries(studyTypeErrors)) {
|
||
if (errorRate > 0.15) {
|
||
issues.push({
|
||
type: 'high_study_type_error_rate',
|
||
studyType,
|
||
errorRate,
|
||
recommendation: `蠅槫刈"${studyType}"邀ｻ蝙狗<EFBFBD>皮ｩｶ逧Ёew-shot譯井ｾ義
|
||
});
|
||
}
|
||
}
|
||
|
||
// 髣ｮ鬚<EFBDAE>3<EFBFBD>夂音螳壽ｨ｡蝙玖｡ｨ邇ｰ蟾ｮ
|
||
const modelErrors = analyzeByModel(sample, humanReviews);
|
||
for (const [model, errorRate] of Object.entries(modelErrors)) {
|
||
if (errorRate > 0.15) {
|
||
issues.push({
|
||
type: 'model_underperformance',
|
||
model,
|
||
errorRate,
|
||
recommendation: `閠<EFBFBD>剔隹<EFBFBD>紛讓｡蝙<EFBFBD>"${model}"逧<>盾謨ｰ謌匁峩謐｢讓｡蝙義
|
||
});
|
||
}
|
||
}
|
||
|
||
return issues;
|
||
}
|
||
```
|
||
|
||
**雍ｨ驥乗冠陦ｨ遉ｺ萓<EFBDBA>**<EFBFBD>?
|
||
|
||
```markdown
|
||
# 蜈ｨ譁<EFBDA8>､咲ｭ幄ｴｨ驥丞ｮ｡隶｡謚･蜻<EFBDA5>
|
||
|
||
**螳｡隶｡蜻ｨ譛<EFBDA8>**<EFBFBD>?025-11-15 閾?2025-11-22
|
||
**諤ｻ謠仙叙謨ｰ**<EFBFBD>?48遽?
|
||
**謚ｽ譬ｷ謨?*<2A>?5遽<35>ｼ<EFBFBD>10.1%<25>?
|
||
|
||
## 謨ｴ菴楢ｴｨ驥乗欠譬<E6ACA0>
|
||
|
||
| 謖<><E8AC96><EFBFBD> | 譛ｬ蜻ｨ | 荳雁捉 | 雜句漢 |
|
||
|------|------|------|------|
|
||
| 蜃<>｡ｮ邇?| 94.7% | 93.2% | 竊?+1.5% |
|
||
| Cohen's Kappa | 0.89 | 0.87 | 竊?+0.02 |
|
||
| 蛛<>亠諤ｧ邇<EFBDA7> | 3.1% | 4.2% | 竊?-1.1% |
|
||
| 蛛<>亢諤ｧ邇<EFBDA7> | 2.2% | 2.6% | 竊?-0.4% |
|
||
|
||
## 蛻<>ｭ玲ｮｵ蜃<EFBDB5>｡ｮ邇<EFBDAE>
|
||
|
||
| 蟄玲ｮｵ | 蜃<>｡ｮ邇?| 迥ｶ諤?|
|
||
|------|--------|------|
|
||
| 遐皮ｩｶ隶ｾ隶｡ | 100% | 笨?莨倡ｧ |
|
||
| 髫乗惻蛹匁婿豕?| 93.3% | 笨?濶ｯ螂ｽ |
|
||
| 逶ｲ豕<EFBDB2> | 86.7% | 笞<><E7AC9E><EFBFBD> 髴謾ｹ霑<EFBDB9> |
|
||
| 蝓ｺ郤ｿ蜿ｯ豈疲?| 100% | 笨?莨倡ｧ |
|
||
| 扈捺棡螳梧紛諤?| 93.3% | 笨?濶ｯ螂ｽ |
|
||
| ... | ... | ... |
|
||
|
||
## 讓｡蝙区ｧ閭ｽ蟇ｹ豈<EFBDB9>
|
||
|
||
| 讓｡蝙<EFBDA1> | 蜃<>｡ｮ邇?| 蟷ｳ蝮<EFBDB3>ｽｮ菫｡蠎?| 螟<>炊譌ｶ髣ｴ |
|
||
|------|--------|-----------|----------|
|
||
| DeepSeek-V3 | 92.1% | 0.87 | 45s |
|
||
| Qwen3-Max | 94.5% | 0.91 | 38s |
|
||
| Claude-4.5<EFBFBD>井ｻｲ陬<EFBFBD>ｼ<EFBFBD> | 97.2% | 0.94 | 62s |
|
||
|
||
## 蜿醍鴫逧<E9B4AB>琉鬚?
|
||
|
||
1. **蟄玲ｮｵ"逶ｲ豕<EFBDB2>"髞呵ｯｯ邇<EFBDAF>￥鬮假ｼ<E58187>13.3%<25>?*
|
||
- 蟶ｸ隗<EFBDB8>漠隸ｯ<E99AB8>壼ｰ<E5A3BC>"蜊慕峇"隸ｯ蛻､荳?螳梧紛"
|
||
- 蜴溷屏蛻<E5B18F>梵<EFBFBD>啀rompt譛ｪ譏守｡ｮ蛹ｺ蛻<EFBDBA>黒逶?蜿檎峇逧<E5B387>ｴｨ驥丞ｷｮ蠑?
|
||
- 謾ｹ霑帛ｻｺ隶ｮ<E99AB6>壽峩譁ｰPrompt<70>悟｢槫<EFBDA2>?蜊慕峇騾壼ｸｸ荳崎ｶｳ莉･髦ｲ豁｢譽豬句￥蛟?逧<>ｯｴ譏?
|
||
|
||
2. **髦溷<E9ABA6>遐皮ｩｶ謠仙叙蜃<E58F99>｡ｮ邇<EFBDAE>ｽ惹ｺ山CT<43>?9% vs 96%<25>?*
|
||
- 蜴溷屏蛻<E5B18F>梵<EFBFBD>夐弌蛻礼<E89BBB>皮ｩｶ逧<EFBDB6>婿豕募ｭｦ謠剰ｿｰ譖ｴ轣ｵ豢ｻ<E8B1A2>梧<EFBFBD><E6A2A7>㊥蛹也ｨ句ｺｦ菴?
|
||
- 謾ｹ霑帛ｻｺ隶ｮ<E99AB6>壼｢槫<EFBDA2>?荳ｪ髦溷<E9ABA6>遐皮ｩｶ逧Ёew-shot譯井ｾ<E4BA95>
|
||
|
||
## 謾ｹ霑帛ｻｺ隶ｮ
|
||
|
||
1. 笨?遶句叉謇ｧ陦鯉ｼ壽峩譁?逶ｲ豕<EFBDB2>"蟄玲ｮｵPrompt讓｡譚ｿ
|
||
2. 笞?譛ｬ蜻ｨ蜀<EFBDA8>ｼ壼｢槫刈髦溷<E9ABA6>遐皮ｩｶFew-shot譯井ｾ句ｺ?
|
||
3. <20>套 荳句捉<E58FA5>夐㍾譁ｰ隸<EFBDB0>ｼ?逶ｲ豕<EFBDB2>"蟄玲ｮｵ蜃<EFBDB5>｡ｮ邇?
|
||
|
||
## 荳句捉逶ｮ譬<EFBDAE>
|
||
|
||
- 蜃<>｡ｮ邇<EFBDAE>ｼ壺<EFBDBC>?95%
|
||
- Cohen's Kappa<70>壺翁 0.90
|
||
- "逶ｲ豕<EFBDB2>"蟄玲ｮｵ蜃<EFBDB5>｡ｮ邇<EFBDAE>ｼ壺<EFBDBC>?93%
|
||
```
|
||
|
||
---
|
||
|
||
#### 1.4 笨?Prompt迚域悽邂｡逅<EFBDA1>
|
||
|
||
**Git邂｡逅<EFBDA1>署遉ｺ隸肴ｨ｡譚ｿ<E8AD9A>梧髪謖、/B豬玖ｯ<E78E96>**<EFBFBD>?
|
||
|
||
```
|
||
backend/prompts/asl/fulltext_screening/
|
||
笏懌楳笏 changelog.md
|
||
笏懌楳笏 fields/
|
||
笏? 笏懌楳笏 髫乗惻蛹匁婿豕?
|
||
笏? 笏? 笏懌楳笏 v1.0.0-basic.md
|
||
笏? 笏? 笏懌楳笏 v1.1.0-with-examples.md
|
||
笏? 笏? 笏懌楳笏 v1.2.0-cot.md
|
||
笏? 笏? 笏披楳笏 v1.3.0-enhanced-cochrane.md 竊?蠖灘燕迚域悽
|
||
笏? 笏懌楳笏 逶ｲ豕<EFBDB2>/
|
||
笏? 笏? 笏懌楳笏 v1.0.0-basic.md
|
||
笏? 笏? 笏懌楳笏 v1.1.0-clarify-single-double.md 竊?謾ｹ霑帷<E99C91>?
|
||
笏? 笏? 笏披楳笏 ...
|
||
笏? 笏披楳笏 ...
|
||
笏披楳笏 tests/
|
||
笏披楳笏 benchmark_results.json
|
||
```
|
||
|
||
**謨ｰ謐ｮ蠎楢ｮｰ蠖?*<2A>?
|
||
|
||
```prisma
|
||
model PromptVersion {
|
||
id String @id @default(uuid())
|
||
|
||
field String // "髫乗惻蛹匁婿豕?
|
||
version String // "v1.3.0"
|
||
content String @db.Text
|
||
changelog String // "蠅槫ｼｺCochrane譬<65>㊥謠剰ｿｰ<EFBDBF>梧ｷｻ蜉?荳ｪFew-shot譯井ｾ<E4BA95>"
|
||
|
||
// 諤ｧ閭ｽ謖<EFBDBD><E8AC96><EFBFBD>ｼ<EFBFBD>/B豬玖ｯ慕ｻ捺棡<E68DBA>?
|
||
accuracy Float? // 0.947
|
||
usageCount Int @default(0)
|
||
avgConfidence Float?
|
||
|
||
// 迥ｶ諤?
|
||
isActive Boolean @default(false)
|
||
isExperimental Boolean @default(false)
|
||
|
||
createdAt DateTime @default(now())
|
||
deactivatedAt DateTime?
|
||
|
||
@@map("asl_prompt_versions")
|
||
}
|
||
```
|
||
|
||
**A/B豬玖ｯ<E78E96>**<EFBFBD>?
|
||
|
||
```typescript
|
||
// 20%豬<>㍼菴ｿ逕ｨ譁ｰ迚<EFBDB0>rompt
|
||
async function extractFieldWithABTest(
|
||
field: string,
|
||
content: string
|
||
) {
|
||
const isExperimentGroup = Math.random() < 0.2;
|
||
|
||
const promptVersion = isExperimentGroup
|
||
? await getPromptVersion(field, 'experimental')
|
||
: await getPromptVersion(field, 'stable');
|
||
|
||
const result = await llmService.chat(
|
||
'deepseek-v3',
|
||
promptVersion.content,
|
||
content
|
||
);
|
||
|
||
// 隶ｰ蠖穂ｽｿ逕ｨ
|
||
await trackPromptUsage({
|
||
field,
|
||
version: promptVersion.version,
|
||
isExperiment: isExperimentGroup,
|
||
result
|
||
});
|
||
|
||
return result;
|
||
}
|
||
|
||
// 豈丞捉蛻<E68D89>梵A/B豬玖ｯ慕ｻ捺棡
|
||
async function analyzeABTest(field: string): Promise<ABTestReport> {
|
||
const stableResults = await getPromptUsageStats(field, 'stable');
|
||
const experimentResults = await getPromptUsageStats(field, 'experimental');
|
||
|
||
const improvement = {
|
||
accuracy: experimentResults.accuracy - stableResults.accuracy,
|
||
confidence: experimentResults.avgConfidence - stableResults.avgConfidence,
|
||
processingTime: experimentResults.avgTime - stableResults.avgTime
|
||
};
|
||
|
||
// 扈溯ｮ｡譏ｾ闡玲ｧ譽鬪?
|
||
const isSignificant = performTTest(stableResults, experimentResults);
|
||
|
||
return {
|
||
field,
|
||
stableVersion: stableResults.version,
|
||
experimentVersion: experimentResults.version,
|
||
sampleSize: {
|
||
stable: stableResults.count,
|
||
experiment: experimentResults.count
|
||
},
|
||
improvement,
|
||
isSignificant,
|
||
recommendation: isSignificant && improvement.accuracy > 0.02
|
||
? 'promote_to_stable' // 謠仙合荳ｺ遞ｳ螳夂沿
|
||
: 'continue_testing' // 扈ｧ扈ｭ豬玖ｯ<E78E96>
|
||
};
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
### 莠後〃2.0謌先悽鬚<EFBFBD>ｮ<EFBFBD>
|
||
|
||
**蝨ｺ譎ｯ<E8AD8E>?00遽<30><E981BD>譁<EFBFBD>､咲ｭ幢ｼ磯ｫ倩ｴｨ驥城｡ｹ逶ｮ<E980B6><EFBDAE>**
|
||
|
||
| 邇ｯ闃<EFBDAF> | Token豸郁?| 讓｡蝙<EFBDA1> | 謌先悽 |
|
||
|------|----------|------|------|
|
||
| 12蟄玲ｮｵ蛻<EFBDB5>ｮｵ謠仙叙<E4BB99>亥曙讓｡蝙具ｼ?| 12K | DeepSeek-V3 + Qwen3-Max | ﾂ･0.06/遽?|
|
||
| 蜈ｨ譁<EFBDA8>ｺ､蜿蛾ｪ瑚ｯ<E7919A> | 3K | DeepSeek-V3 | ﾂ･0.003/遽?|
|
||
| 蜈ｳ髞ｮ蟄玲ｮｵ荳画ｨ｡蝙倶ｻｲ陬<EFBDB2>ｼ<EFBFBD>15%<25>?| 3K | Claude-4.5 | ﾂ･0.03/遽<>ｼ井ｻ?5%<25>?|
|
||
| 雍ｨ驥丞ｮ｡隶｡<E99AB6>?0%謚ｽ譟･<E8AD9F>?| 2K | 莠ｺ蟾･ | 10蛻<30>帖/遽?|
|
||
| **100遽<30>ｻ謌先<E8AC8C>?* | - | - | **ﾂ･10 + 莠ｺ蟾･謌先悽** |
|
||
|
||
**雍ｨ驥乗署蜊<E7BDB2>**<EFBFBD>壼㊥遑ｮ邇<EFBFBD> 92% 竊?96%
|
||
**謌先悽蠅槫刈**<EFBFBD>堋? 竊?ﾂ･10<31>?25%<25>御ｽ<E5BEA1>ｾｾ蛻ｰ蛹ｻ蟄ｦ郤ｧ譬<EFBDA7>㊥<EFBFBD><E38AA5>
|
||
|
||
---
|
||
|
||
### 荳峨〃2.0鬪梧噺譬<EFBFBD>㊥
|
||
|
||
| 謖<><E8AC96><EFBFBD> | 逶ｮ譬<EFBDAE> | 鬪瑚ｯ∵婿豕<E5A9BF> |
|
||
|------|------|----------|
|
||
| 蜃<>｡ｮ邇<EFBDAE>ｼ井ｸ灘ｮｶ隸<EFBDB6>ｼｰ<EFBDBC>?| 竕?96% | 莠ｺ蟾･謚ｽ譟･100遽?|
|
||
| 莠ｺ譛ｺ荳閾ｴ諤?| Cohen's Kappa 竕?0.90 | 扈溯ｮ｡蛻<EFBDA1>梵 |
|
||
| 蛛<>亠諤ｧ邇<EFBDA7> | 竕?3% | 扈溯ｮ｡蛻<EFBDA1>梵 |
|
||
| 蛛<>亢諤ｧ邇<EFBDA7> | 竕?2% | 扈溯ｮ｡蛻<EFBDA1>梵 |
|
||
| 隸∵紺體ｾ螳梧紛諤?| 100% | 閾ｪ蜉ｨ譽譟?|
|
||
| 閾ｪ蜉ｨ蛹門ｮ｡隶?| 豈丞捉1谺?| 邉ｻ扈滓冠陦ｨ |
|
||
| Prompt迚域悽邂｡逅<EFBDA1> | 100% | Git蜴<74>彰霑ｽ雕ｪ |
|
||
| 隨ｦ蜷<EFBDA6>ochrane譬<65>㊥ | 竕?95% | 荳灘ｮｶ隶､隸<EFBDA4> |
|
||
|
||
---
|
||
|
||
## <20>投 荳蛾亳谿ｵ蟇ｹ豈疲ｻ扈<EFBDBB>
|
||
|
||
| 扈ｴ蠎ｦ | MVP | V1.0 | V2.0 |
|
||
|------|-----|------|------|
|
||
| **蜃<>｡ｮ邇?* | 85% | 92% | 96% |
|
||
| **譬ｸ蠢<EFBDB8>ｭ也払** | Nougat+蛻<>ｮｵ謠仙叙 | +蜈ｨ譁<EFBDA8>ｪ瑚ｯ<E7919A>+騾ｻ霎題ｧ<E9A18C><EFBDA7> | +荳画ｨ｡蝙倶ｻｲ陬?螳｡隶｡ |
|
||
| **隸∵紺體?* | 蝓ｺ譛ｬ蠑慕畑 | 螳梧紛螳壻ｽ<E5A3BB> | 螳｡隶｡郤ｧ譌･蠢?|
|
||
| **雍ｨ驥乗而蛻ｶ** | 蜿梧ｨ｡蝙矩ｪ瑚ｯ?| 蛹ｻ蟄ｦ騾ｻ霎大ｼ墓梼 | HITL+閾ｪ蜉ｨ螳｡隶｡ |
|
||
| **謌先悽/100遽?* | ﾂ･6 | ﾂ･8 | ﾂ･10 |
|
||
| **蠑蜿大捉譛?* | 3蜻?| 5蜻?| 8蜻?|
|
||
| **騾ら畑蝨ｺ譎ｯ** | 蠢ｫ騾滄ｪ瑚ｯ?| 蟶ｸ隗<EFBDB8>｡ｹ逶ｮ | Cochrane蜿題｡ｨ |
|
||
|
||
---
|
||
|
||
## <20>売 螳樊命霍ｯ蠕<EFBDAF>
|
||
|
||
### 髦ｶ谿ｵ1<EFBDB5>哺VP蠑蜿托ｼ<E68998>eek 1-3<>?
|
||
|
||
**Week 1**<EFBFBD>壼渕遑譫ｶ譫<EFBFBD>
|
||
- [x] PDF蟄伜お譛榊苅<E6A68A>亥ｷｲ螳梧<E89EB3><E6A2A7>俄怛
|
||
- [ ] Nougat謠仙叙+遶<>闃りｧ｣譫<EFBDA3>
|
||
- [ ] 12蟄玲ｮｵ霍ｯ逕ｱ陦ｨ隶ｾ隶?
|
||
- [ ] 蝓ｺ遑Prompt讓｡譚ｿ<E8AD9A>?2荳ｪ蟄玲ｮｵ<EFBDAE><EFBDB5>
|
||
|
||
**Week 2**<EFBFBD>壽<EFBFBD>ｸ蠢<EFBFBD>粥閭?
|
||
- [ ] 蛻<>ｮｵ蟷ｶ陦梧署蜿<E7BDB2>
|
||
- [ ] 蜿梧ｨ｡蝙玖ｰ<E78E96><EFBDB0>?
|
||
- [ ] 蟄玲ｮｵ郤ｧ蜀ｲ遯∵｣豬?
|
||
- [ ] 蝓ｺ遑隸∵紺體?
|
||
|
||
**Week 3**<EFBFBD>壼燕遶?豬玖ｯ<E78E96>
|
||
- [ ] 蜑咲ｫｯ蟾･菴懷<E88FB4>?
|
||
- [ ] 蜀ｲ遯∝ｯｹ豈碑ｧ<E7A291>崟
|
||
- [ ] 莠ｺ蟾･螟肴<E89E9F>ｸ逡碁擇
|
||
- [ ] 蜉溯<E89C89>豬玖ｯ<E78E96>+蜃<>｡ｮ邇<EFBDAE>ｯ<EFBFBD>ｼ?
|
||
|
||
### 髦ｶ谿ｵ2<EFBDB5>啖1.0蠅槫ｼｺ<EFBFBD><EFBFBD>eek 4-8<>?
|
||
|
||
**Week 4-5**<EFBFBD>夊ｴｨ驥乗署蜊?
|
||
- [ ] Cochrane譬<65>㊥Prompt蠅槫ｼｺ
|
||
- [ ] Few-shot蛹ｻ蟄ｦ譯井ｾ句ｺ難ｼ域ｯ丞ｭ玲ｮ?-5荳ｪ<E88DB3><EFBDAA>
|
||
- [ ] CoT謗ｨ逅<EFBDA8>｢槫ｼｺ
|
||
|
||
**Week 6-7**<EFBFBD>夐ｪ瑚ｯ∵惻蛻?
|
||
- [ ] 蜈ｨ譁<EFBDA8>ｺ､蜿蛾ｪ瑚ｯ<E7919A>
|
||
- [ ] 蛹ｻ蟄ｦ騾ｻ霎題ｧ<E9A18C><EFBDA7>蠑墓梼
|
||
- [ ] 螳梧紛隸∵紺體?
|
||
|
||
**Week 8**<EFBFBD>壻ｼ伜<EFBFBD>?譁<>｡｣
|
||
- [ ] 諤ｧ閭ｽ莨伜喧
|
||
- [ ] A/B豬玖ｯ<E78E96>
|
||
- [ ] 譁<>｡｣螳悟埋
|
||
|
||
### 髦ｶ谿ｵ3<EFBDB5>啖2.0螳悟埋<EFBFBD><EFBFBD>eek 9-16<31>?
|
||
|
||
**Week 9-11**<EFBFBD>夐ｫ倡ｺｧ蜉溯<EFBFBD>?
|
||
- [ ] 荳画ｨ｡蝙倶ｻｲ陬?
|
||
- [ ] HITL譎ｺ閭ｽ蛻<EFBDBD>ｵ<EFBFBD>
|
||
- [ ] Prompt迚域悽邂｡逅<EFBDA1>+A/B豬玖ｯ<E78E96>
|
||
|
||
**Week 12-14**<EFBFBD>夊ｴｨ驥丞ｮ｡隶?
|
||
- [ ] 閾ｪ蜉ｨ螳｡隶｡邉ｻ扈<EFBDBB>
|
||
- [ ] 雍ｨ驥乗冠陦ｨ
|
||
- [ ] 蠑ょｸｸ譽豬?
|
||
|
||
**Week 15-16**<EFBFBD>壼現蟄ｦ荳灘ｮｶ鬪瑚ｯ?
|
||
- [ ] Cochrane荳灘ｮｶ隸<EFBDB6>ｮ｡
|
||
- [ ] 蜈ｨ驥乗ｵ玖ｯ<E78E96>
|
||
- [ ] 蜿大ｸ<E5A4A7>枚譯｣
|
||
|
||
---
|
||
|
||
## <20>答 逶ｸ蜈ｳ譁<EFBDB3>｡｣
|
||
|
||
- [譬<EFBFBD>｢俶遭隕∝<EFBFBD>遲幄ｴｨ驥丈ｿ晞囿遲也払](./06-雍ｨ驥丈ｿ晞囿荳主庄霑ｽ貅ｯ遲也払.md)
|
||
- [蜈ｨ譁<EFBDA8>､咲ｭ帛ｼ蜿題ｮ｡蛻綻(../04-蠑蜿題ｮ｡蛻?04-蜈ｨ譁<EFBDA8>､咲ｭ帛ｼ蜿題ｮ｡蛻?md)
|
||
- [謨ｰ謐ｮ蠎楢ｮｾ隶｡](./01-謨ｰ謐ｮ蠎楢ｮｾ隶?md)
|
||
- [API隶ｾ隶｡隗<EFBFBD>激](./02-API隶ｾ隶｡隗<EFBDA1>激.md)
|
||
- [莠大次逕溷ｼ蜿題ｧ<EFBFBD>激](../../../04-蠑蜿題ｧ<E9A18C><EFBDA7>?08-莠大次逕溷ｼ蜿題ｧ<E9A18C><EFBDA7>?md)
|
||
|
||
---
|
||
|
||
**譖ｴ譁ｰ譌･蠢<EFBDA5>**<EFBFBD>?
|
||
- 2025-11-22: 蛻帛ｻｺ譁<EFBDBA>｡｣<EFBDA1>悟ｮ壻ｹ牙<EFBDB9>譁<EFBFBD>､咲ｭ帑ｸ蛾亳谿ｵ雍ｨ驥丈ｿ晞囿遲也払
|
||
- 蝓ｺ莠晒ougat扈捺桷蛹?蛻<>ｮｵ謠仙叙+蜈ｨ譁<EFBDA8>ｪ瑚ｯ∫噪謚譛ｯ譁ｹ譯?
|
||
- 蜿りイochrane RoB 2.0譬<EFBFBD>㊥隶ｾ隶｡荳謎ｸ啀rompt讓｡譚ｿ
|
||
- 蠑ｺ隹<EFBDBA>ｮ梧紛隸∵紺體ｾ蜥悟庄霑ｽ貅ｯ諤?
|
||
|
||
|
||
|