feat: 优化足迹总结与自身消息强度分析
This commit is contained in:
clearyss committed
1 parent
ca6c479496
commit
a67959dc2a
8 files changed
+503
-53
No files matched your search
@@ -41,6 +41,18 @@ const API_MAX_TOKENS_MIN = 1
|
||||
const API_MAX_TOKENS_MAX = 2_000_000
|
||||
const API_TEMPERATURE = 0.7
|
||||
const INSIGHT_NOTIFICATION_AVATAR_URL = './assets/insight/AI_Insight.png'
|
||||
const MIMO_FOOTPRINT_MIN_TOKENS = 4096
|
||||
const FOOTPRINT_API_TEMPERATURE = 0.2
|
||||
|
||||
const DEFAULT_FOOTPRINT_SYSTEM_PROMPT = `你是“我的微信足迹”模块的总结器，只能根据用户提供的统计数据生成最终复盘文案。
|
||||
硬性输出规则：
|
||||
1. 只输出最终总结正文，不输出思考过程、步骤、标题、列表、JSON、Markdown、代码块、引号或字段名。
|
||||
2. 输出 2 句中文，总长度 60-160 字，最多 180 字。
|
||||
3. 第 1 句概括联络活跃度、回复情况或 @我情况；第 2 句给出一个当天/当前范围内可执行的沟通建议。
|
||||
4. 必须引用至少 2 个输入数字，例如人数、回复率、@我次数或群聊数。
|
||||
5. 数据为 0 时如实说明，不臆测具体聊天内容、关系、情绪、诊断或原因。
|
||||
6. 禁止出现“首先”“其次”“根据”“综上”“作为AI”“我认为”“以下是”等过程性表达。
|
||||
输出格式：直接输出两句自然中文。`
|
||||
|
||||
/** 沉默天数阈值默认值 */
|
||||
const DEFAULT_SILENCE_DAYS = 3
|
||||
@@ -81,6 +93,12 @@ interface SharedAiModelConfig {
|
||||
|
||||
type InsightFilterMode = 'whitelist' | 'blacklist'
|
||||
|
||||
interface CallApiOptions {
|
||||
temperature?: number
|
||||
disableThinking?: boolean
|
||||
useMaxCompletionTokens?: boolean
|
||||
}
|
||||
|
||||
// ─── 日志 ─────────────────────────────────────────────────────────────────────
|
||||
|
||||
type InsightLogLevel = 'INFO' | 'WARN' | 'ERROR'
|
||||
@@ -161,6 +179,54 @@ function normalizeSessionIdList(value: unknown): string[] {
|
||||
return Array.from(new Set(value.map((item) => String(item || '').trim()).filter(Boolean)))
|
||||
}
|
||||
|
||||
function isMimoModel(apiBaseUrl: string, model: string): boolean {
|
||||
const target = `${apiBaseUrl} ${model}`.toLowerCase()
|
||||
return target.includes('mimo') || target.includes('xiaomi')
|
||||
}
|
||||
|
||||
function buildFootprintSystemPrompt(customPrompt: string): string {
|
||||
const custom = String(customPrompt || '').trim()
|
||||
if (!custom || custom === DEFAULT_FOOTPRINT_SYSTEM_PROMPT) {
|
||||
return DEFAULT_FOOTPRINT_SYSTEM_PROMPT
|
||||
}
|
||||
return `${DEFAULT_FOOTPRINT_SYSTEM_PROMPT}
|
||||
|
||||
用户自定义补充要求如下，只能在不违反上述硬性输出规则时执行：
|
||||
${custom}`
|
||||
}
|
||||
|
||||
function normalizeFootprintInsight(text: string): string {
|
||||
let normalized = String(text || '').trim()
|
||||
if (!normalized) return ''
|
||||
|
||||
if (normalized.startsWith('{') && normalized.endsWith('}')) {
|
||||
try {
|
||||
const parsed = JSON.parse(normalized)
|
||||
const value = parsed?.summary || parsed?.insight || parsed?.content || parsed?.text
|
||||
if (typeof value === 'string' && value.trim()) {
|
||||
normalized = value.trim()
|
||||
}
|
||||
} catch { }
|
||||
}
|
||||
|
||||
normalized = normalized
|
||||
.replace(/^```(?:text|markdown|md|json)?/i, '')
|
||||
.replace(/```$/i, '')
|
||||
.replace(/^(足迹复盘|AI足迹总结|AI 足迹总结|总结|建议)[:：]\s*/i, '')
|
||||
.replace(/^\s*[-*•]\s*/gm, '')
|
||||
.replace(/\s*\n+\s*/g, ' ')
|
||||
.replace(/\s{2,}/g, ' ')
|
||||
.trim()
|
||||
|
||||
if (normalized.length > 180) {
|
||||
const sliced = normalized.slice(0, 180)
|
||||
const lastStop = Math.max(sliced.lastIndexOf('。'), sliced.lastIndexOf('！'), sliced.lastIndexOf('？'))
|
||||
normalized = lastStop >= 60 ? sliced.slice(0, lastStop + 1) : `${sliced.replace(/[，,；;、\s]+$/g, '')}。`
|
||||
}
|
||||
|
||||
return normalized
|
||||
}
|
||||
|
||||
/**
|
||||
* 调用 OpenAI 兼容 API（非流式），返回模型第一条消息内容。
|
||||
* 使用 Node 原生 https/http 模块，无需任何第三方 SDK。
|
||||
@@ -171,7 +237,8 @@ function callApi(
|
||||
model: string,
|
||||
messages: Array<{ role: string; content: string }>,
|
||||
timeoutMs: number = API_TIMEOUT_MS,
|
||||
maxTokens: number = API_MAX_TOKENS_DEFAULT
|
||||
maxTokens: number = API_MAX_TOKENS_DEFAULT,
|
||||
options: CallApiOptions = {}
|
||||
): Promise<string> {
|
||||
return new Promise((resolve, reject) => {
|
||||
const endpoint = buildApiUrl(apiBaseUrl, '/chat/completions')
|
||||
@@ -183,15 +250,26 @@ function callApi(
|
||||
return
|
||||
}
|
||||
|
||||
const body = JSON.stringify({
|
||||
const normalizedMaxTokens = normalizeApiMaxTokens(maxTokens)
|
||||
const requestBody: Record<string, unknown> = {
|
||||
model,
|
||||
messages,
|
||||
max_tokens: normalizeApiMaxTokens(maxTokens),
|
||||
temperature: API_TEMPERATURE,
|
||||
temperature: options.temperature ?? API_TEMPERATURE,
|
||||
stream: false
|
||||
})
|
||||
}
|
||||
if (options.useMaxCompletionTokens) {
|
||||
requestBody.max_completion_tokens = normalizedMaxTokens
|
||||
} else {
|
||||
requestBody.max_tokens = normalizedMaxTokens
|
||||
}
|
||||
if (options.disableThinking) {
|
||||
requestBody.thinking = { type: 'disabled' }
|
||||
requestBody.enable_thinking = false
|
||||
}
|
||||
|
||||
const options = {
|
||||
const body = JSON.stringify(requestBody)
|
||||
|
||||
const requestOptions = {
|
||||
hostname: urlObj.hostname,
|
||||
port: urlObj.port || (urlObj.protocol === 'https:' ? 443 : 80),
|
||||
path: urlObj.pathname + urlObj.search,
|
||||
@@ -205,7 +283,7 @@ function callApi(
|
||||
|
||||
const isHttps = urlObj.protocol === 'https:'
|
||||
const requestFn = isHttps ? https.request : http.request
|
||||
const req = requestFn(options, (res) => {
|
||||
const req = requestFn(requestOptions, (res) => {
|
||||
let data = ''
|
||||
res.on('data', (chunk) => { data += chunk })
|
||||
res.on('end', () => {
|
||||
@@ -215,7 +293,13 @@ function callApi(
|
||||
if (typeof content === 'string' && content.trim()) {
|
||||
resolve(content.trim())
|
||||
} else {
|
||||
reject(new Error(`API 返回格式异常: ${data.slice(0, 200)}`))
|
||||
const finishReason = parsed?.choices?.[0]?.finish_reason
|
||||
const reasoningContent = parsed?.choices?.[0]?.message?.reasoning_content
|
||||
if (typeof reasoningContent === 'string' && reasoningContent.trim()) {
|
||||
reject(new Error(`API 仅返回推理内容未返回正文${finishReason ? `（finish_reason=${finishReason}）` : ''}，请增大最大输出 Token 或关闭思考模式`))
|
||||
return
|
||||
}
|
||||
reject(new Error(`API 返回格式异常${finishReason ? `（finish_reason=${finishReason}）` : ''}: ${data.slice(0, 200)}`))
|
||||
}
|
||||
} catch (e) {
|
||||
reject(new Error(`JSON 解析失败: ${data.slice(0, 200)}`))
|
||||
@@ -523,6 +607,7 @@ class InsightService {
|
||||
const rangeLabel = String(params?.rangeLabel || '').trim() || '当前范围'
|
||||
const privateSegments = Array.isArray(params?.privateSegments) ? params.privateSegments.slice(0, 6) : []
|
||||
const mentionGroups = Array.isArray(params?.mentionGroups) ? params.mentionGroups.slice(0, 6) : []
|
||||
const mimoMode = isMimoModel(apiBaseUrl, model)
|
||||
|
||||
const topPrivateText = privateSegments.length > 0
|
||||
? privateSegments
|
||||
@@ -546,20 +631,31 @@ class InsightService {
|
||||
.join('\n')
|
||||
: '无'
|
||||
|
||||
const defaultSystemPrompt = `你是用户的聊天足迹教练，负责基于统计数据给出一段简明复盘。
|
||||
要求：
|
||||
1. 输出 2-3 句，总长度不超过 180 字。
|
||||
2. 必须包含：总体观察 + 一个可执行建议。
|
||||
3. 语气务实，不夸张，不使用 Markdown。`
|
||||
const customPrompt = String(this.config.get('aiFootprintSystemPrompt') || '').trim()
|
||||
const systemPrompt = customPrompt || defaultSystemPrompt
|
||||
const systemPrompt = buildFootprintSystemPrompt(customPrompt)
|
||||
|
||||
const userPromptBase = `统计范围：${rangeLabel}
|
||||
有聊天的人数：${Number(summary.private_inbound_people) || 0}
|
||||
我有回复的人数：${Number(summary.private_outbound_people) || 0}
|
||||
回复率：${(((Number(summary.private_reply_rate) || 0) * 100)).toFixed(1)}%
|
||||
@我次数：${Number(summary.mention_count) || 0}
|
||||
涉及群聊：${Number(summary.mention_group_count) || 0}
|
||||
const inboundPeople = Number(summary.private_inbound_people) || 0
|
||||
const repliedPeople = Number(summary.private_replied_people) || 0
|
||||
const outboundPeople = Number(summary.private_outbound_people) || 0
|
||||
const replyRate = (((Number(summary.private_reply_rate) || 0) * 100)).toFixed(1)
|
||||
const mentionCount = Number(summary.mention_count) || 0
|
||||
const mentionGroupCount = Number(summary.mention_group_count) || 0
|
||||
|
||||
const userPromptBase = `任务：基于下面的“我的微信足迹”统计生成最终总结正文。
|
||||
|
||||
输出要求再强调一次：
|
||||
- 只输出 2 句中文自然语言，不要输出分析过程。
|
||||
- 不要输出 JSON / Markdown / 列表 / 标题 / 代码块。
|
||||
- 第 1 句做总体观察，第 2 句给一个可执行建议。
|
||||
- 必须引用至少 2 个统计数字。
|
||||
|
||||
统计范围：${rangeLabel}
|
||||
有聊天的人数：${inboundPeople}
|
||||
我有回复的人数：${outboundPeople}
|
||||
实际回复了其中：${repliedPeople}
|
||||
回复率：${replyRate}%
|
||||
@我次数：${mentionCount}
|
||||
涉及群聊：${mentionGroupCount}
|
||||
|
||||
私聊重点：
|
||||
${topPrivateText}
|
||||
@@ -567,7 +663,7 @@ ${topPrivateText}
|
||||
群聊@我重点：
|
||||
${topMentionText}
|
||||
|
||||
请给出足迹复盘（2-3句，含建议）：`
|
||||
现在直接输出最终总结正文：`
|
||||
const userPrompt = appendPromptCurrentTime(userPromptBase)
|
||||
|
||||
try {
|
||||
@@ -580,9 +676,14 @@ ${topMentionText}
|
||||
{ role: 'user', content: userPrompt }
|
||||
],
|
||||
25_000,
|
||||
maxTokens
|
||||
mimoMode ? Math.max(maxTokens, MIMO_FOOTPRINT_MIN_TOKENS) : maxTokens,
|
||||
{
|
||||
temperature: FOOTPRINT_API_TEMPERATURE,
|
||||
disableThinking: mimoMode,
|
||||
useMaxCompletionTokens: mimoMode
|
||||
}
|
||||
)
|
||||
const insight = result.trim()
|
||||
const insight = normalizeFootprintInsight(result)
|
||||
if (!insight) return { success: false, message: '模型返回为空' }
|
||||
return { success: true, message: '生成成功', insight }
|
||||
} catch (error) {
|
||||
@@ -1329,5 +1430,3 @@ ${topMentionText}
|
||||
}
|
||||
|
||||
export const insightService = new InsightService()
|
||||
|
||||
|
||||
Reference in new issue
Block a user