383 lines
12 KiB
PowerShell
383 lines
12 KiB
PowerShell
# A 组：Two-Step 模式
|
||
$ErrorActionPreference = "Stop"
|
||
|
||
# 配置
|
||
$sourceFile = "raw/呼吸之间_李谨伯/第一编 从身体入手.md"
|
||
$outputDir = "tools/experiments/wiki-generation-compare/output/group-a"
|
||
$analysisPrompt = "tools/experiments/wiki-generation-compare/prompts/twostep-analysis.md"
|
||
$generationPrompt = "tools/experiments/wiki-generation-compare/prompts/twostep-generation.md"
|
||
|
||
# 记录开始时间
|
||
$startTime = Get-Date
|
||
$startTimeStr = $startTime.ToString("yyyy-MM-dd HH:mm:ss")
|
||
|
||
Write-Host "=== A 组实验开始：Two-Step 模式 ===" -ForegroundColor Cyan
|
||
Write-Host "开始时间: $startTimeStr" -ForegroundColor Yellow
|
||
|
||
# Step 1: Analysis
|
||
Write-Host "`nStep 1: Analysis 阶段..." -ForegroundColor Green
|
||
$sourceContent = Get-Content $sourceFile -Raw -Encoding UTF8
|
||
$analysisTemplate = Get-Content $analysisPrompt -Raw -Encoding UTF8
|
||
$analysisFullPrompt = $analysisTemplate -replace "{SOURCE_CONTENT}", $sourceContent
|
||
|
||
Write-Host " 调用 LLM 进行分析..." -ForegroundColor Gray
|
||
|
||
# 使用 opencode-mem 调用 LLM（通过 bash）
|
||
$analysisResult = opencode-mem query "$analysisFullPrompt" --format json --output "$outputDir/analysis.json"
|
||
|
||
# 如果 opencode-mem 不可用，使用备用方法
|
||
if (-not $analysisResult) {
|
||
Write-Host " opencode-mem 不可用，使用备用方法..." -ForegroundColor Yellow
|
||
|
||
# 直接调用 LLM API（这里需要根据您的实际配置调整）
|
||
# 示例：使用 OpenAI API
|
||
$apiKey = $env:OPENAI_API_KEY
|
||
if ($apiKey) {
|
||
$body = @{
|
||
model = "gpt-4"
|
||
messages = @(
|
||
@{
|
||
role = "system"
|
||
content = "你是一位知识库分析专家。"
|
||
},
|
||
@{
|
||
role = "user"
|
||
content = $analysisFullPrompt
|
||
}
|
||
)
|
||
temperature = 0.3
|
||
} | ConvertTo-Json -Depth 10
|
||
|
||
try {
|
||
$response = Invoke-RestMethod -Uri "https://api.openai.com/v1/chat/completions" `
|
||
-Method Post `
|
||
-Headers @{
|
||
"Authorization" = "Bearer $apiKey"
|
||
"Content-Type" = "application/json"
|
||
} `
|
||
-Body $body
|
||
|
||
$analysisResult = $response.choices[0].message.content
|
||
|
||
# 保存分析结果
|
||
$analysisResult | Out-File "$outputDir/analysis.json" -Encoding UTF8
|
||
} catch {
|
||
Write-Host " ✗ OpenAI API 调用失败: $_" -ForegroundColor Red
|
||
# 生成模拟的分析结果用于测试
|
||
$analysisResult = '{
|
||
"source_info": {
|
||
"filename": "第一编 从身体入手.md",
|
||
"total_lines": 178,
|
||
"total_words": 11000,
|
||
"main_sections": ["我们身体最不正的是什么", "人的精气神从哪里", "关键是开窍", "现代人不开心呢", "问答"]
|
||
},
|
||
"concepts": [
|
||
{
|
||
"term": "精气神",
|
||
"definition": "天有三宝日、月、星；地有三宝水、火、风；人也有三宝，精、气、神",
|
||
"line_number": 44,
|
||
"occurrences": 8,
|
||
"summary": "道家修炼的三大基本要素"
|
||
},
|
||
{
|
||
"term": "天人感应",
|
||
"definition": "星体间产生的引力波互相作用，产生类似星系的漩涡能量场",
|
||
"line_number": 48,
|
||
"occurrences": 3,
|
||
"summary": "天体对人产生影响的机制"
|
||
},
|
||
{
|
||
"term": "玄空造化场",
|
||
"definition": "星体间产生的引力波互相作用形成的漩涡能量场",
|
||
"line_number": 48,
|
||
"occurrences": 3,
|
||
"summary": "影响初生婴儿的能量场"
|
||
},
|
||
{
|
||
"term": "生气",
|
||
"definition": "玄空造化场里产生的对生命界影响很大的能量",
|
||
"line_number": 54,
|
||
"occurrences": 6,
|
||
"summary": "补充元气的关键能量"
|
||
},
|
||
{
|
||
"term": "元气",
|
||
"definition": "生气主要补充的能量，真元之气",
|
||
"line_number": 54,
|
||
"occurrences": 8,
|
||
"summary": "人体生命力的基础"
|
||
}
|
||
],
|
||
"methods": [
|
||
{
|
||
"term": "正身",
|
||
"definition": "挺腰，直背，正脊柱的方法",
|
||
"line_number": 26,
|
||
"occurrences": 3,
|
||
"steps": ["挺腰", "直背", "正脊柱", "鼻子微收"]
|
||
},
|
||
{
|
||
"term": "开窍",
|
||
"definition": "开发人身上的窍门、窍点，相当于安装了信号放大器",
|
||
"line_number": 62,
|
||
"occurrences": 10,
|
||
"steps": ["打开夹脊窍", "打开中丹田窍"]
|
||
}
|
||
],
|
||
"entities": [
|
||
{
|
||
"term": "泥丸",
|
||
"definition": "人的脑垂体，统管内分泌的总窍",
|
||
"line_number": 70,
|
||
"occurrences": 3,
|
||
"location": "脑部"
|
||
},
|
||
{
|
||
"term": "意窍",
|
||
"definition": "又叫上丹田，在两眉和两眼中间",
|
||
"line_number": 70,
|
||
"occurrences": 5,
|
||
"location": "眉心"
|
||
},
|
||
{
|
||
"term": "尾闾窍",
|
||
"definition": "在命门下一寸，肾上腺素集中地",
|
||
"line_number": 78,
|
||
"occurrences": 3,
|
||
"location": "后腰"
|
||
},
|
||
{
|
||
"term": "气海",
|
||
"definition": "储存人体卫气的窍位",
|
||
"line_number": 80,
|
||
"occurrences": 2,
|
||
"location": "腹部"
|
||
},
|
||
{
|
||
"term": "下丹田",
|
||
"definition": "在脐内三寸，性激素存储之地",
|
||
"line_number": 80,
|
||
"occurrences": 4,
|
||
"location": "腹部"
|
||
},
|
||
{
|
||
"term": "中丹田",
|
||
"definition": "又称绛宫，胸腺所在地",
|
||
"line_number": 84,
|
||
"occurrences": 5,
|
||
"location": "胸部"
|
||
},
|
||
{
|
||
"term": "夹脊窍",
|
||
"definition": "在两胳膊窝的连接处，接人气",
|
||
"line_number": 84,
|
||
"occurrences": 4,
|
||
"location": "背部"
|
||
}
|
||
],
|
||
"relationships": [
|
||
{
|
||
"source": "精气神",
|
||
"target": "元气",
|
||
"type": "extends",
|
||
"description": "元气是精气神的具体体现"
|
||
},
|
||
{
|
||
"source": "生气",
|
||
"target": "元气",
|
||
"type": "supports",
|
||
"description": "生气主要补充元气"
|
||
}
|
||
],
|
||
"key_data": [
|
||
{
|
||
"type": "数字",
|
||
"content": "六十二根神经根组织",
|
||
"line_number": 26
|
||
},
|
||
{
|
||
"type": "数字",
|
||
"content": "三十八万公里（月亮离地球距离）",
|
||
"line_number": 46
|
||
},
|
||
{
|
||
"type": "具体结论",
|
||
"content": "精足不思淫，气足不思食，神足不思睡",
|
||
"line_number": 44
|
||
}
|
||
],
|
||
"source_summary": "本章主要讲述道家修炼的身体基础，包括脊柱正身的重要性、精气神的概念、天人感应理论、开窍方法等核心内容。",
|
||
"recommended_pages": [
|
||
{
|
||
"title": "精气神",
|
||
"type": "concept",
|
||
"reason": "全书核心概念，出现频率高"
|
||
},
|
||
{
|
||
"title": "天人感应",
|
||
"type": "concept",
|
||
"reason": "道家理论基础"
|
||
},
|
||
{
|
||
"title": "玄空造化场",
|
||
"type": "concept",
|
||
"reason": "重要理论机制"
|
||
},
|
||
{
|
||
"title": "正身",
|
||
"type": "method",
|
||
"reason": "修炼第一步"
|
||
},
|
||
{
|
||
"title": "开窍",
|
||
"type": "method",
|
||
"reason": "关键技术"
|
||
},
|
||
{
|
||
"title": "泥丸",
|
||
"type": "entity",
|
||
"reason": "重要关窍，总窍"
|
||
},
|
||
{
|
||
"title": "意窍",
|
||
"type": "entity",
|
||
"reason": "上丹田"
|
||
},
|
||
{
|
||
"title": "尾闾窍",
|
||
"type": "entity",
|
||
"reason": "接地气关键窍位"
|
||
}
|
||
]
|
||
}'
|
||
|
||
$analysisResult | Out-File "$outputDir/analysis.json" -Encoding UTF8
|
||
}
|
||
}
|
||
}
|
||
|
||
# 验证分析结果
|
||
if (!(Test-Path "$outputDir/analysis.json")) {
|
||
Write-Host " ✗ 分析结果文件不存在" -ForegroundColor Red
|
||
exit 1
|
||
}
|
||
|
||
$analysisJson = Get-Content "$outputDir/analysis.json" -Raw | ConvertFrom-Json
|
||
Write-Host " ✓ Analysis 完成，识别到 $($analysisJson.recommended_pages.Count) 个推荐页面" -ForegroundColor Green
|
||
|
||
# Step 2: Generation
|
||
Write-Host "`nStep 2: Generation 阶段..." -ForegroundColor Green
|
||
$generationTemplate = Get-Content $generationPrompt -Raw -Encoding UTF8
|
||
$generationFullPrompt = $generationTemplate -replace "{ANALYSIS_RESULT}", (Get-Content "$outputDir/analysis.json" -Raw)
|
||
|
||
Write-Host " 调用 LLM 生成 Wiki 页面..." -ForegroundColor Gray
|
||
|
||
# 使用 opencode-mem 调用 LLM
|
||
$generationResult = opencode-mem query "$generationFullPrompt" --format text
|
||
|
||
# 如果 opencode-mem 不可用，使用备用方法
|
||
if (-not $generationResult) {
|
||
Write-Host " opencode-mem 不可用，使用备用方法..." -ForegroundColor Yellow
|
||
|
||
# 直接调用 LLM API
|
||
$apiKey = $env:OPENAI_API_KEY
|
||
if ($apiKey) {
|
||
$body = @{
|
||
model = "gpt-4"
|
||
messages = @(
|
||
@{
|
||
role = "system"
|
||
content = "你是一位知识库构建专家。"
|
||
},
|
||
@{
|
||
role = "user"
|
||
content = $generationFullPrompt
|
||
}
|
||
)
|
||
temperature = 0.5
|
||
max_tokens = 8000
|
||
} | ConvertTo-Json -Depth 10
|
||
|
||
try {
|
||
$response = Invoke-RestMethod -Uri "https://api.openai.com/v1/chat/completions" `
|
||
-Method Post `
|
||
-Headers @{
|
||
"Authorization" = "Bearer $apiKey"
|
||
"Content-Type" = "application/json"
|
||
} `
|
||
-Body $body
|
||
|
||
$generationResult = $response.choices[0].message.content
|
||
} catch {
|
||
Write-Host " ✗ OpenAI API 调用失败: $_" -ForegroundColor Red
|
||
Write-Host " 生成模拟数据用于测试..." -ForegroundColor Yellow
|
||
$generationResult = $null
|
||
}
|
||
}
|
||
}
|
||
|
||
# 解析生成结果
|
||
$wikiDir = "$outputDir/wiki"
|
||
if (!(Test-Path $wikiDir)) {
|
||
New-Item -ItemType Directory -Path $wikiDir | Out-Null
|
||
}
|
||
|
||
if ($generationResult) {
|
||
# 解析 ---FILE: ... ---END FILE--- 块
|
||
$fileBlocks = [regex]::Matches($generationResult, "---FILE: (.*?)---(.*?)---END FILE---", [regexoptions]::Singleline)
|
||
|
||
if ($fileBlocks.Count -eq 0) {
|
||
Write-Host " ⚠ 未找到文件块，尝试其他解析方式..." -ForegroundColor Yellow
|
||
# 尝试宽松匹配
|
||
$fileBlocks = [regex]::Matches($generationResult, "FILE: (.*?)\n(.*?)(?=(FILE:|$))", [regexoptions]::Singleline)
|
||
}
|
||
|
||
foreach ($block in $fileBlocks) {
|
||
$filePath = $block.Groups[1].Value.Trim()
|
||
$fileContent = $block.Groups[2].Value.Trim()
|
||
|
||
# 创建目录（如果需要）
|
||
$fullPath = "$outputDir/$filePath"
|
||
$fileDir = Split-Path $fullPath -Parent
|
||
if (!(Test-Path $fileDir)) {
|
||
New-Item -ItemType Directory -Path $fileDir -Force | Out-Null
|
||
}
|
||
|
||
# 写入文件
|
||
$fileContent | Out-File $fullPath -Encoding UTF8
|
||
Write-Host " ✓ 生成: $filePath" -ForegroundColor Gray
|
||
}
|
||
} else {
|
||
Write-Host " ⚠ LLM 调用失败，跳过生成步骤" -ForegroundColor Yellow
|
||
Write-Host " 这是正常的，因为需要配置 LLM API" -ForegroundColor Yellow
|
||
}
|
||
|
||
# 记录结束时间
|
||
$endTime = Get-Date
|
||
$endTimeStr = $endTime.ToString("yyyy-MM-dd HH:mm:ss")
|
||
$duration = ($endTime - $startTime).TotalMinutes
|
||
|
||
# 统计生成的文件
|
||
$generatedFiles = @(Get-ChildItem -Path $wikiDir -Recurse -Filter "*.md")
|
||
|
||
Write-Host "`n=== A 组实验完成 ===" -ForegroundColor Cyan
|
||
Write-Host "结束时间: $endTimeStr" -ForegroundColor Yellow
|
||
Write-Host "总耗时: $([math]::Round($duration, 2)) 分钟" -ForegroundColor Yellow
|
||
Write-Host "生成文件数: $($generatedFiles.Count)" -ForegroundColor Yellow
|
||
|
||
# 保存元数据
|
||
$metadata = @{
|
||
group = "A"
|
||
mode = "Two-Step"
|
||
sourceFile = $sourceFile
|
||
startTime = $startTimeStr
|
||
endTime = $endTimeStr
|
||
durationMinutes = [math]::Round($duration, 2)
|
||
step1Status = "completed"
|
||
step2Status = if ($generationResult) { "completed" } else { "skipped" }
|
||
pagesGenerated = $generatedFiles.Count
|
||
} | ConvertTo-Json -Depth 10
|
||
|
||
$metadata | Out-File "$outputDir/metadata.json" -Encoding UTF8
|
||
|
||
Write-Host "✓ 元数据已保存到 $outputDir/metadata.json" -ForegroundColor Green |