feat(lead): 一键发布增强 — 物料清单提取 + 标题预校验 + 错误诊断
- 新增 MaterialExtractorService：发布前实时从公告正文/标题提取结构化物料(items)， 三层降级(正文 AI → 标题 AI → 整标题兜底)，全程不阻断发布 - publishToDirev 接入物料提取，items 传给产业平台 publish_rfq - 新增 B2B 标题 50 字预校验，提前报错避免拿到通用"服务器内部异常" - 新增 diagnosePublishError()：把产业平台原始错误翻译成明确指引 (标题超长/数量异常/身份校验/必填缺失)
This commit is contained in:
Sisyphus
parent
a822d7782b
commit
8a3c96598b
@ -4,13 +4,14 @@ import { IntegrationModule } from '../integration/integration.module'
|
||||
import { CrawlerModule } from '../crawler/crawler.module'
|
||||
import { ProductCategoryModule } from '../product-category/product-category.module'
|
||||
import { LeadService } from './lead.service'
|
||||
import { MaterialExtractorService } from './material-extractor.service'
|
||||
import { LeadController } from './lead.controller'
|
||||
import { LeadNotesController } from './lead-notes.controller'
|
||||
|
||||
@Module({
|
||||
imports: [PrismaModule, IntegrationModule, CrawlerModule, ProductCategoryModule],
|
||||
controllers: [LeadController, LeadNotesController],
|
||||
providers: [LeadService],
|
||||
providers: [LeadService, MaterialExtractorService],
|
||||
exports: [LeadService],
|
||||
})
|
||||
export class LeadModule {}
|
||||
|
||||
@ -3,6 +3,7 @@ import { PrismaService } from '../prisma/prisma.service'
|
||||
import { IntegrationService } from '../integration/integration.service'
|
||||
import { ProductCategoryService } from '../product-category/product-category.service'
|
||||
import { CategoryInferenceService } from '../product-category/category-inference.service'
|
||||
import { MaterialExtractorService } from './material-extractor.service'
|
||||
|
||||
@Injectable()
|
||||
export class LeadService {
|
||||
@ -13,6 +14,7 @@ export class LeadService {
|
||||
private integrationService: IntegrationService,
|
||||
private categoryService: ProductCategoryService,
|
||||
private categoryInference: CategoryInferenceService,
|
||||
private materialExtractor: MaterialExtractorService,
|
||||
) {}
|
||||
|
||||
async findAll(params?: { status?: string; level?: string; search?: string; star?: string; source?: string; category?: string; assignedTo?: string; timeRange?: string }) {
|
||||
@ -308,6 +310,32 @@ export class LeadService {
|
||||
publishArgs.enterpriseName = `${b2bIntegration.name}公司`
|
||||
}
|
||||
|
||||
// 预校验: B2B 平台字段长度限制 (tb_inquiry.title varchar(50))
|
||||
// 在调 MCP 之前先报错，避免拿到通用的"服务器内部异常"
|
||||
const TITLE_MAX_LEN = 50
|
||||
if (publishArgs.title && String(publishArgs.title).length > TITLE_MAX_LEN) {
|
||||
const titleLen = String(publishArgs.title).length
|
||||
throw new ServiceUnavailableException(
|
||||
[
|
||||
`发布失败: 询价标题超过 B2B 平台限制(最多 ${TITLE_MAX_LEN} 字)。`,
|
||||
`当前标题长度: ${titleLen} 字`,
|
||||
`建议: ① 把标题精简到 ${TITLE_MAX_LEN} 字以内后再发布; ② 或联系 B2B 平台把 tb_inquiry.title 字段改为 varchar(200) 永久解决。`,
|
||||
`线索标题: ${String(publishArgs.title).slice(0, 40)}${titleLen > 40 ? '...' : ''}`,
|
||||
].join('\n'),
|
||||
)
|
||||
}
|
||||
|
||||
// 实时提取物料清单（detailUrl 正文 → 标题 → 兜底），传给产业平台 items 字段
|
||||
try {
|
||||
const items = await this.materialExtractor.extractItems(lead)
|
||||
if (items.length > 0) {
|
||||
publishArgs.items = items
|
||||
this.logger.log(`[发布] 物料清单提取完成 leadId=${id} ${items.length} 行`)
|
||||
}
|
||||
} catch (e: any) {
|
||||
this.logger.warn(`[发布] 物料清单提取失败(不阻断发布) leadId=${id}: ${e.message}`)
|
||||
}
|
||||
|
||||
const result = await this.integrationService.executeTool<any>(
|
||||
b2bIntegration.id,
|
||||
'publish_rfq',
|
||||
@ -331,9 +359,9 @@ export class LeadService {
|
||||
const payload = data.data || data
|
||||
const rfqId = payload.rfqId || payload.id || null
|
||||
if (!rfqId) {
|
||||
throw new ServiceUnavailableException(
|
||||
`发布失败: B2B 平台返回了响应但未包含询价 ID。响应: ${JSON.stringify(payload).slice(0, 200)}`,
|
||||
)
|
||||
// 识别常见错误模式，给出明确提示
|
||||
const diagnosis = this.diagnosePublishError(lead, payload)
|
||||
throw new ServiceUnavailableException(diagnosis)
|
||||
}
|
||||
const inquiryNo = payload.orderId || rfqId
|
||||
const direvUrl = `${b2bIntegration.endpoint}/inquiry/${rfqId}`
|
||||
@ -443,4 +471,54 @@ export class LeadService {
|
||||
return []
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* 发布失败的诊断提示 — 把产业平台原始错误翻译成明确指引
|
||||
* 输入: lead(知道当前 title/quantity 等) + payload(产业平台响应)
|
||||
* 输出: 给前端展示的友好错误信息
|
||||
*/
|
||||
private diagnosePublishError(
|
||||
lead: {
|
||||
title?: string | null
|
||||
quantity?: number | null
|
||||
region?: string | null
|
||||
deadline?: string | null
|
||||
},
|
||||
payload: any,
|
||||
): string {
|
||||
// 把 payload 序列化用于关键词匹配
|
||||
const raw = JSON.stringify(payload || {}).slice(0, 500)
|
||||
const msg = String(payload?.msg || payload?.message || '')
|
||||
|
||||
// 1. 标题超长（产业平台 tb_inquiry.title 是 varchar(50)）
|
||||
if (
|
||||
/Data truncation|Data too long.*title|column 'title'.*too long/i.test(raw + ' ' + msg)
|
||||
) {
|
||||
const titleLen = (lead.title || '').length
|
||||
return [
|
||||
'发布失败: 询价标题超过 B2B 平台限制(最多 50 字)。',
|
||||
`当前线索标题长度: ${titleLen} 字`,
|
||||
`建议: ① 把标题精简到 50 字以内; ② 或联系 B2B 平台把 tb_inquiry.title 字段改为 varchar(200)。`,
|
||||
`线索标题: ${(lead.title || '').slice(0, 30)}${titleLen > 30 ? '...' : ''}`,
|
||||
].join('\n')
|
||||
}
|
||||
|
||||
// 2. 数量字段异常
|
||||
if (/Data too long.*qty|column 'qty'|column 'quantity'/i.test(raw + ' ' + msg)) {
|
||||
return `发布失败: 询价数量字段超过 B2B 平台限制。当前 quantity=${lead.quantity}。建议改成合理数字(≤1000)。`
|
||||
}
|
||||
|
||||
// 3. 企业/用户身份未通过
|
||||
if (/enterprise|user.*id|companyUserId|身份|权限/i.test(msg)) {
|
||||
return `发布失败: B2B 平台校验企业身份未通过(${msg})。请检查运营管理 → 系统配置 → MCP 对接配置的连接状态(状态灯应为绿色"已连接")。`
|
||||
}
|
||||
|
||||
// 4. 必填字段缺失
|
||||
if (/required|不能为空|not.*null|缺失/i.test(msg)) {
|
||||
return `发布失败: B2B 平台提示必填字段缺失(${msg})。线索当前数据：quantity=${lead.quantity ?? '空'} region=${lead.region || '空'} deadline=${lead.deadline || '空'}。请补充完整后再发布。`
|
||||
}
|
||||
|
||||
// 5. 默认：展示原始错误
|
||||
return `发布失败: B2B 平台返回了响应但未包含询价 ID。原因: ${msg || '未知错误'}。响应: ${raw}`
|
||||
}
|
||||
}
|
||||
|
||||
280
apps/api/src/lead/material-extractor.service.ts
Normal file
280
apps/api/src/lead/material-extractor.service.ts
Normal file
@ -0,0 +1,280 @@
|
||||
import { Injectable, Logger } from '@nestjs/common'
|
||||
|
||||
const MINIMAX_ENDPOINT = 'https://api.minimaxi.com/v1/chat/completions'
|
||||
const MINIMAX_MODEL = process.env.MINIMAX_MODEL || 'MiniMax-M2.7'
|
||||
|
||||
const DETAIL_FETCH_TIMEOUT_MS = 10_000 // detailUrl 正文抓取超时
|
||||
const TEXT_EXTRACT_TIMEOUT_MS = 30_000 // 正文物料提取超时
|
||||
const TITLE_EXTRACT_TIMEOUT_MS = 15_000 // 标题物资名提取超时
|
||||
const MAX_INPUT_CHARS = 4000
|
||||
|
||||
export interface MaterialItem {
|
||||
productName: string
|
||||
quantity: number
|
||||
unit: string
|
||||
specification?: string
|
||||
brand?: string
|
||||
note?: string
|
||||
}
|
||||
|
||||
const TEXT_SYSTEM_PROMPT = `你是招标公告物资清单提取助手。从以下公告内容中提取物资/物料明细。
|
||||
|
||||
输出格式：纯 JSON 数组，不要解释，不要 markdown 代码块，不要 <think> 标签。
|
||||
每项格式：{"productName":"物资名称","quantity":数字,"unit":"单位","specification":"规格型号"}
|
||||
|
||||
规则：
|
||||
- productName 必填，物资的具体名称（如：电力电缆、水泵、变压器）
|
||||
- quantity 提取到的数量，没有就填 1
|
||||
- unit 计量单位（件/米/吨/台/批等），没有就填"批"
|
||||
- specification 规格型号，没有就填空字符串
|
||||
- 最多输出 20 行
|
||||
- 如果内容里没有具体物资清单，返回 []`
|
||||
|
||||
const TITLE_SYSTEM_PROMPT = `你是物资名称提取助手。从招标公告标题中提取涉及的物资名称。
|
||||
|
||||
输出格式：纯 JSON 数组，不要解释，不要 markdown 代码块，不要 <think> 标签。
|
||||
每项格式：{"productName":"物资名称"}
|
||||
|
||||
规则：
|
||||
- 只提取具体的物资/设备名称（如：油浸式电力变压器、电压互感器、水泵、逆变器）
|
||||
- 去掉"采购""招标""项目""变更公告"等噪声词
|
||||
- 如果标题里有多个物资（用顿号/逗号分隔），每个都提取
|
||||
- 如果标题只提到一类物资，提取 1 条
|
||||
- 如果完全无法判断物资名称，返回 []`
|
||||
|
||||
/**
|
||||
* 物料清单提取服务
|
||||
*
|
||||
* 发布到产业平台时实时提取，降级链：
|
||||
* 1. 抓 detailUrl 正文 → AI 提取多行物料
|
||||
* 2. 抓不到 → AI 从标题提取物资名 (quantity=1, unit="批")
|
||||
* 3. AI 也失败 → 用 lead.title 兜底
|
||||
*/
|
||||
@Injectable()
|
||||
export class MaterialExtractorService {
|
||||
private readonly logger = new Logger(MaterialExtractorService.name)
|
||||
|
||||
/**
|
||||
* 主入口 - 从 Lead 提取物料清单
|
||||
* 全程降级，任何异常都返回至少 1 条兜底数据，不阻断发布
|
||||
*/
|
||||
async extractItems(lead: { id: string; title: string; detailUrl?: string | null }): Promise<MaterialItem[]> {
|
||||
const apiKey = process.env.MINIMAX_API_KEY
|
||||
if (!apiKey) {
|
||||
this.logger.warn(`[物料提取] MINIMAX_API_KEY 未配置，走标题兜底 leadId=${lead.id}`)
|
||||
return this.fallbackWithTitle(lead.title)
|
||||
}
|
||||
|
||||
// 第 1 步：尝试抓 detailUrl 正文
|
||||
if (lead.detailUrl) {
|
||||
try {
|
||||
const text = await this.fetchDetailText(lead.detailUrl)
|
||||
if (text.length > 100) {
|
||||
this.logger.log(`[物料提取] 抓到正文 leadId=${lead.id} textLen=${text.length}`)
|
||||
const items = await this.extractFromText(text, apiKey)
|
||||
if (items.length > 0) {
|
||||
this.logger.log(`[物料提取] 正文提取成功 leadId=${lead.id} ${items.length} 行`)
|
||||
return items
|
||||
}
|
||||
this.logger.log(`[物料提取] 正文提取为空，降级标题 leadId=${lead.id}`)
|
||||
} else {
|
||||
this.logger.log(`[物料提取] 正文过短(${text.length}字)，降级标题 leadId=${lead.id}`)
|
||||
}
|
||||
} catch (e: any) {
|
||||
this.logger.warn(`[物料提取] 抓取正文失败 leadId=${lead.id}: ${e.message}`)
|
||||
}
|
||||
}
|
||||
|
||||
// 第 2 步：从标题提取
|
||||
try {
|
||||
const items = await this.extractFromTitle(lead.title, apiKey)
|
||||
if (items.length > 0) {
|
||||
this.logger.log(`[物料提取] 标题提取成功 leadId=${lead.id} ${items.length} 行`)
|
||||
return items
|
||||
}
|
||||
} catch (e: any) {
|
||||
this.logger.warn(`[物料提取] 标题提取失败 leadId=${lead.id}: ${e.message}`)
|
||||
}
|
||||
|
||||
// 第 3 步：兜底
|
||||
this.logger.log(`[物料提取] 走标题兜底 leadId=${lead.id}`)
|
||||
return this.fallbackWithTitle(lead.title)
|
||||
}
|
||||
|
||||
/** 抓 detailUrl HTML 并转纯文本，失败返回空串 */
|
||||
private async fetchDetailText(detailUrl: string): Promise<string> {
|
||||
const r = await fetch(detailUrl, {
|
||||
signal: AbortSignal.timeout(DETAIL_FETCH_TIMEOUT_MS),
|
||||
headers: { 'User-Agent': 'Mozilla/5.0 (iPhone; CPU iPhone OS 17_0 like Mac OS X)' },
|
||||
})
|
||||
if (!r.ok) throw new Error(`HTTP ${r.status}`)
|
||||
const html = await r.text()
|
||||
return this.htmlToText(html)
|
||||
}
|
||||
|
||||
/** 从正文提取物料行 */
|
||||
private async extractFromText(text: string, apiKey: string): Promise<MaterialItem[]> {
|
||||
const truncated = text.length > MAX_INPUT_CHARS
|
||||
? text.slice(0, MAX_INPUT_CHARS) + '\n[内容过长已截断]'
|
||||
: text
|
||||
|
||||
const content = await this.callMinimax(
|
||||
TEXT_SYSTEM_PROMPT,
|
||||
truncated,
|
||||
apiKey,
|
||||
TEXT_EXTRACT_TIMEOUT_MS,
|
||||
2000,
|
||||
)
|
||||
return this.parseItems(content)
|
||||
}
|
||||
|
||||
/** 从标题提取物资名，自动补 quantity=1 unit=批 */
|
||||
private async extractFromTitle(title: string, apiKey: string): Promise<MaterialItem[]> {
|
||||
const content = await this.callMinimax(
|
||||
TITLE_SYSTEM_PROMPT,
|
||||
title,
|
||||
apiKey,
|
||||
TITLE_EXTRACT_TIMEOUT_MS,
|
||||
800,
|
||||
)
|
||||
|
||||
// 解析 [{productName: "xxx"}, ...]
|
||||
const raw = this.extractJsonArray(content)
|
||||
if (!raw) return []
|
||||
let parsed: any[]
|
||||
try {
|
||||
parsed = JSON.parse(raw)
|
||||
} catch {
|
||||
return []
|
||||
}
|
||||
|
||||
return parsed
|
||||
.filter((item: any) => item && typeof item.productName === 'string' && item.productName.trim())
|
||||
.map((item: any) => ({
|
||||
productName: String(item.productName).trim(),
|
||||
quantity: 1,
|
||||
unit: '批',
|
||||
specification: '',
|
||||
}))
|
||||
}
|
||||
|
||||
/** 调 MiniMax chat completion，返回 choices[0].message.content */
|
||||
private async callMinimax(
|
||||
systemPrompt: string,
|
||||
userContent: string,
|
||||
apiKey: string,
|
||||
timeoutMs: number,
|
||||
maxTokens: number,
|
||||
): Promise<string> {
|
||||
const r = await fetch(MINIMAX_ENDPOINT, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
Authorization: `Bearer ${apiKey}`,
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
body: JSON.stringify({
|
||||
model: MINIMAX_MODEL,
|
||||
messages: [
|
||||
{ role: 'system', content: systemPrompt },
|
||||
{ role: 'user', content: userContent },
|
||||
],
|
||||
temperature: 0,
|
||||
max_tokens: maxTokens,
|
||||
}),
|
||||
signal: AbortSignal.timeout(timeoutMs),
|
||||
})
|
||||
|
||||
if (!r.ok) {
|
||||
const errText = await r.text().catch(() => '')
|
||||
throw new Error(`MiniMax API ${r.status}: ${errText.slice(0, 200)}`)
|
||||
}
|
||||
|
||||
const data = (await r.json()) as { choices?: Array<{ message?: { content?: string } }> }
|
||||
const content = data.choices?.[0]?.message?.content || ''
|
||||
if (!content) throw new Error(`MiniMax 返回空: ${JSON.stringify(data).slice(0, 200)}`)
|
||||
return content
|
||||
}
|
||||
|
||||
/** 从 MiniMax 响应解析物料数组，带脏数据清洗 + 字段校验 */
|
||||
private parseItems(content: string): MaterialItem[] {
|
||||
const jsonStr = this.extractJsonArray(content)
|
||||
if (!jsonStr) return []
|
||||
|
||||
let parsed: any[]
|
||||
try {
|
||||
parsed = JSON.parse(jsonStr)
|
||||
} catch {
|
||||
return []
|
||||
}
|
||||
if (!Array.isArray(parsed)) return []
|
||||
|
||||
const items: MaterialItem[] = []
|
||||
for (const raw of parsed) {
|
||||
if (!raw || typeof raw !== 'object') continue
|
||||
const productName = typeof raw.productName === 'string' ? raw.productName.trim() : ''
|
||||
if (!productName) continue
|
||||
|
||||
// quantity 必须是正数
|
||||
let quantity = 1
|
||||
if (typeof raw.quantity === 'number' && raw.quantity > 0) {
|
||||
quantity = raw.quantity
|
||||
} else if (typeof raw.quantity === 'string') {
|
||||
const n = parseFloat(raw.quantity)
|
||||
if (!isNaN(n) && n > 0) quantity = n
|
||||
}
|
||||
|
||||
const unit = typeof raw.unit === 'string' && raw.unit.trim() ? raw.unit.trim() : '批'
|
||||
const specification = typeof raw.specification === 'string' ? raw.specification.trim() : ''
|
||||
|
||||
items.push({ productName, quantity, unit, specification })
|
||||
if (items.length >= 20) break
|
||||
}
|
||||
return items
|
||||
}
|
||||
|
||||
/**
|
||||
* 从模型输出中提取 JSON 数组字符串
|
||||
* 处理 <think> 标签、markdown 代码块、前后缀解释文字
|
||||
*/
|
||||
private extractJsonArray(content: string): string | null {
|
||||
// 1. 剥 <think>...</think> 和 <think> 开放标签
|
||||
let s = content
|
||||
.replace(/<\/?think>/g, '')
|
||||
.replace(/\*\*/g, '')
|
||||
.trim()
|
||||
|
||||
// 2. 尝试从 markdown 代码块提取 ```json ... ``` 或 ``` ... ```
|
||||
const codeBlock = s.match(/```(?:json)?\s*([\s\S]*?)```/i)
|
||||
if (codeBlock) s = codeBlock[1].trim()
|
||||
|
||||
// 3. 提取第一个 [...] JSON 数组
|
||||
const match = s.match(/\[[\s\S]*\]/)
|
||||
return match ? match[0] : null
|
||||
}
|
||||
|
||||
/** 兜底：用标题当物资名 */
|
||||
private fallbackWithTitle(title: string): MaterialItem[] {
|
||||
return [{ productName: title || '未命名物资', quantity: 1, unit: '批', specification: '' }]
|
||||
}
|
||||
|
||||
/** HTML 转纯文本（复用 notes-extractor 的清洗逻辑） */
|
||||
private htmlToText(html: string): string {
|
||||
return html
|
||||
.replace(/<script\b[^<]*(?:(?!<\/script>)<[^<]*)*<\/script>/gi, ' ')
|
||||
.replace(/<style\b[^<]*(?:(?!<\/style>)<[^<]*)*<\/style>/gi, ' ')
|
||||
.replace(/<noscript\b[^<]*(?:(?!<\/noscript>)<[^<]*)*<\/noscript>/gi, ' ')
|
||||
.replace(/<head\b[^<]*(?:(?!<\/head>)<[^<]*)*<\/head>/gi, ' ')
|
||||
.replace(/<!--[\s\S]*?-->/g, ' ')
|
||||
.replace(/<[^>]+>/g, '\n')
|
||||
.replace(/ /g, ' ')
|
||||
.replace(/&/g, '&')
|
||||
.replace(/</g, '<')
|
||||
.replace(/>/g, '>')
|
||||
.replace(/"/g, '"')
|
||||
.replace(/&#(\d+);/g, (_, dec) => String.fromCharCode(Number(dec)))
|
||||
.split('\n')
|
||||
.map(l => l.trim())
|
||||
.filter(l => l.length > 0)
|
||||
.join('\n')
|
||||
}
|
||||
}
|
||||
Loading…
Reference in New Issue
Block a user