mirror of
https://wget.la/https://github.com/Wxw-Gu/WechatExplorer
synced 2026-08-18 03:57:02 +08:00
feat: 完善群聊日报模板与图片理解
This commit is contained in:
@@ -0,0 +1,94 @@
|
||||
// src/main/services/image-insight-prompt.ts
|
||||
// 图片理解 prompt 模板 — 输出严格的 JSON,便于程序化解析
|
||||
|
||||
export const IMAGE_ANALYSIS_SYSTEM_PROMPT = `你是微信群聊的图片分析助手。
|
||||
请根据用户提供的图片和图片前后的聊天上下文,生成对该图片的结构化理解。
|
||||
|
||||
输出要求(严格遵守):
|
||||
1. 必须输出 JSON,不要用 markdown 代码块包裹
|
||||
2. description:1-2 句中文,30-80 字,描述图片核心内容
|
||||
3. ocrText:如果图片含文字(截图、文档、票据等),提取出来;纯风景/表情包可填空字符串
|
||||
4. tags:3-6 个中文关键词标签
|
||||
5. category:screenshot / photo / meme / document / chart / other 之一
|
||||
6. importance:low / medium / high — 根据图片的信息密度和后续讨论热度判断
|
||||
|
||||
禁止:
|
||||
- 不要猜测图片中未明确可见的内容
|
||||
- 不要复述聊天上下文本身(那是 description 之外的事)
|
||||
- 不要输出 markdown 标记`
|
||||
|
||||
export interface ImageAnalysisContext {
|
||||
sender: string
|
||||
sentAt: number
|
||||
contextBefore: string[] // 图片前 1-3 条消息
|
||||
contextAfter: string[] // 图片后 1-3 条消息
|
||||
}
|
||||
|
||||
export function buildImageAnalysisUserText(ctx: ImageAnalysisContext): string {
|
||||
const before = ctx.contextBefore.length
|
||||
? ctx.contextBefore.map((m, i) => ` ${i + 1}. ${m}`).join('\n')
|
||||
: ' (无前文)'
|
||||
const after = ctx.contextAfter.length
|
||||
? ctx.contextAfter.map((m, i) => ` ${i + 1}. ${m}`).join('\n')
|
||||
: ' (无后续讨论)'
|
||||
const time = new Date(ctx.sentAt * 1000).toLocaleString('zh-CN', { hour12: false })
|
||||
|
||||
return `发送者:${ctx.sender}
|
||||
时间:${time}
|
||||
|
||||
图片前的聊天:
|
||||
${before}
|
||||
|
||||
图片后的聊天:
|
||||
${after}
|
||||
|
||||
请输出 JSON(严格遵守 system 要求):
|
||||
{"description":"...","ocrText":"...","tags":["..."],"category":"...","importance":"..."}`
|
||||
}
|
||||
|
||||
/**
|
||||
* 把 AI 文本响应解析成结构化字段。
|
||||
* 容忍:无 markdown 包裹、有 markdown 包裹、尾部有杂质等。
|
||||
*/
|
||||
export function parseImageAnalysisResponse(raw: string): {
|
||||
description: string
|
||||
ocrText: string
|
||||
tags: string[]
|
||||
category: 'screenshot' | 'photo' | 'meme' | 'document' | 'chart' | 'other'
|
||||
importance: 'low' | 'medium' | 'high'
|
||||
} {
|
||||
const text = raw.trim()
|
||||
// 提取 JSON 段
|
||||
const jsonMatch = text.match(/\{[\s\S]*\}/)
|
||||
if (!jsonMatch) {
|
||||
throw new Error('AI 未返回合法 JSON')
|
||||
}
|
||||
let parsed: Record<string, unknown>
|
||||
try {
|
||||
parsed = JSON.parse(jsonMatch[0])
|
||||
} catch {
|
||||
throw new Error('AI 返回的 JSON 无法解析')
|
||||
}
|
||||
|
||||
const description = String(parsed.description || '').trim()
|
||||
if (!description) throw new Error('AI 未返回 description')
|
||||
|
||||
const ocrText = String(parsed.ocrText || '').trim()
|
||||
const tagsRaw = parsed.tags
|
||||
const tags = Array.isArray(tagsRaw)
|
||||
? tagsRaw.map((t) => String(t).trim()).filter(Boolean).slice(0, 8)
|
||||
: []
|
||||
|
||||
const categoryRaw = String(parsed.category || 'other').toLowerCase()
|
||||
const category: 'screenshot' | 'photo' | 'meme' | 'document' | 'chart' | 'other' =
|
||||
['screenshot', 'photo', 'meme', 'document', 'chart'].includes(categoryRaw)
|
||||
? (categoryRaw as 'screenshot' | 'photo' | 'meme' | 'document' | 'chart')
|
||||
: 'other'
|
||||
|
||||
const importanceRaw = String(parsed.importance || 'medium').toLowerCase()
|
||||
const importance: 'low' | 'medium' | 'high' = ['low', 'medium', 'high'].includes(importanceRaw)
|
||||
? (importanceRaw as 'low' | 'medium' | 'high')
|
||||
: 'medium'
|
||||
|
||||
return { description, ocrText, tags, category, importance }
|
||||
}
|
||||
Reference in New Issue
Block a user