158 lines
6.4 KiB
Python
158 lines
6.4 KiB
Python
|
"""补一条"r1到r5分别代表什么"的 FAQ 知识（幂等；解决该问法答不出来的问题）。
|
|||
|
|
|
|||
|
|
## 为什么需要它
|
|||
|
|
|
|||
|
|
用户实测：问「r1到r5分别代表什么」时客服不给答案（走兜底 + 转人工）。排查结论（2026-09-14）：
|
|||
|
|
|
|||
|
|
- 向量库**有**内容，检索也命中（top1 = 那一套"高频问答对"里的 `R1到R5风险等级是什么意思?`，
|
|||
|
|
相似度 **0.6621**）；
|
|||
|
|
- 但客服的判定是"中置信（0.55–0.75）必须**领先次优 ≥0.07**"（
|
|||
|
|
`app/service/agent/implementations/customer_service.py:148-150`），而这一问的次优是
|
|||
|
|
0.6412 → **只领先 0.021** → 判"候选并列"，转人工。
|
|||
|
|
|
|||
|
|
根因是**同一主题有多个来源**：`高频问答对.txt` 的 R1–R5 条目、《个人投资者适当性管理指南》
|
|||
|
|
第十一条（还切成好几块）、产品手册的 1.4 节——它们分数天然挤在一起，gap 永远拉不开。
|
|||
|
|
|
|||
|
|
## 这条为什么有效
|
|||
|
|
|
|||
|
|
实测（本机，qwen3-embedding）：
|
|||
|
|
|
|||
|
|
| 条目形态 | 「r1到r5分别代表什么」的 top1 | 结论 |
|
|||
|
|
|---|---|---|
|
|||
|
|
| 长条目（表格 + 详细说明，488+433 字，切两块） | 0.6622 | 向量被长正文稀释，仍不过门槛 |
|
|||
|
|
| 短条目（272 字，标题与问法近似） | 0.7214 | gap 0.059，仍差一点 |
|
|||
|
|
| **短条目 + 标题与问法逐字对齐**（本条的做法） | **0.7384**，次优 0.6623 → **gap 0.0761 ≥ 0.07** | ✅ **可答** |
|
|||
|
|
|
|||
|
|
也就是：**命中条目的标题/正文与客户问法越贴近、块越短，相似度越高**；高置信档（≥0.75）
|
|||
|
|
不要求 gap，中置信档靠"领先次优"过关。内容侧这一条就能修好，不必改门槛策略。
|
|||
|
|
|
|||
|
|
## 用法
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
# 先看要做什么（默认只读探测）
|
|||
|
|
python tools/seed_knowledge_r1r5_faq.py
|
|||
|
|
|
|||
|
|
# 真正上传（需要 API 在跑，且 Worker 在跑以便投向量）
|
|||
|
|
python tools/seed_knowledge_r1r5_faq.py --apply
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
幂等：若已存在同 `source_file` 的**未过期**知识，直接跳过（要覆盖先加 `--force`，
|
|||
|
|
它会先删旧的再传新的）。
|
|||
|
|
"""
|
|||
|
|
|
|||
|
|
from __future__ import annotations
|
|||
|
|
|
|||
|
|
import argparse
|
|||
|
|
import asyncio
|
|||
|
|
import base64
|
|||
|
|
import sys
|
|||
|
|
import time
|
|||
|
|
import uuid
|
|||
|
|
from pathlib import Path
|
|||
|
|
|
|||
|
|
import httpx
|
|||
|
|
|
|||
|
|
if hasattr(sys.stdout, "reconfigure"):
|
|||
|
|
sys.stdout.reconfigure(errors="replace")
|
|||
|
|
|
|||
|
|
BASE_DEFAULT = "http://127.0.0.1:8000"
|
|||
|
|
SOURCE = Path(__file__).resolve().parent.parent / "data" / "knowledge" / "faq_r1r5.md"
|
|||
|
|
FILENAME = "r1到r5分别代表什么.md"
|
|||
|
|
PROBE_QUERY = "r1到r5分别代表什么"
|
|||
|
|
|
|||
|
|
|
|||
|
|
def _login(client: httpx.Client, base: str, username: str, password: str) -> str:
|
|||
|
|
response = client.post(
|
|||
|
|
f"{base}/api/v1/auth/tokens", json={"username": username, "password": password}
|
|||
|
|
)
|
|||
|
|
if response.status_code != 200:
|
|||
|
|
raise SystemExit(f"登录失败：HTTP {response.status_code} {response.text[:160]}")
|
|||
|
|
return str(response.json()["data"]["access_token"])
|
|||
|
|
|
|||
|
|
|
|||
|
|
def _headers(token: str) -> dict[str, str]:
|
|||
|
|
return {"Authorization": f"Bearer {token}", "Idempotency-Key": uuid.uuid4().hex}
|
|||
|
|
|
|||
|
|
|
|||
|
|
def _existing_active(client: httpx.Client, base: str, token: str) -> list[dict[str, object]]:
|
|||
|
|
"""列未过期知识，挑出同 source_file 的条目（用于幂等）。"""
|
|||
|
|
response = client.get(f"{base}/api/v1/knowledge/list", headers=_headers(token))
|
|||
|
|
body = response.json()
|
|||
|
|
items = body.get("items")
|
|||
|
|
if items is None:
|
|||
|
|
data = body.get("data")
|
|||
|
|
items = data.get("items") if isinstance(data, dict) else data
|
|||
|
|
if not isinstance(items, list):
|
|||
|
|
return []
|
|||
|
|
return [row for row in items if isinstance(row, dict) and row.get("source_file") == FILENAME]
|
|||
|
|
|
|||
|
|
|
|||
|
|
def _probe_scores() -> None:
|
|||
|
|
"""上传后核对检索评分（与客服判定同一口径）。"""
|
|||
|
|
from app.service.agent.bootstrap import get_knowledge_search_service
|
|||
|
|
|
|||
|
|
service = get_knowledge_search_service()
|
|||
|
|
outcome = asyncio.run(service.search(PROBE_QUERY, top_k=5))
|
|||
|
|
if not outcome.hits:
|
|||
|
|
print(f" ⚠️ 「{PROBE_QUERY}」仍然没有命中")
|
|||
|
|
return
|
|||
|
|
top = outcome.hits[0].score
|
|||
|
|
second = outcome.hits[1].score if len(outcome.hits) > 1 else 0.0
|
|||
|
|
gap = top - second
|
|||
|
|
verdict = "高置信直接答" if top >= 0.75 else (
|
|||
|
|
f"中置信 gap={gap:.4f} → " + ("可答" if gap >= 0.07 else "仍会转人工")
|
|||
|
|
)
|
|||
|
|
print(f" 检索：top1={top:.4f}（{outcome.hits[0].title[:30]}） 次优={second:.4f} → {verdict}")
|
|||
|
|
|
|||
|
|
|
|||
|
|
def main() -> int:
|
|||
|
|
parser = argparse.ArgumentParser(description="补 R1–R5 的 FAQ 知识（幂等）")
|
|||
|
|
parser.add_argument("--base", default=BASE_DEFAULT, help="平台地址")
|
|||
|
|
parser.add_argument("--username", default="admin_t")
|
|||
|
|
parser.add_argument("--password", default="88888888")
|
|||
|
|
parser.add_argument("--apply", action="store_true", help="真正上传（默认只看要做什么）")
|
|||
|
|
parser.add_argument("--force", action="store_true", help="已存在时删旧的再传")
|
|||
|
|
args = parser.parse_args()
|
|||
|
|
|
|||
|
|
if not SOURCE.exists():
|
|||
|
|
raise SystemExit(f"知识正文不存在：{SOURCE}")
|
|||
|
|
content = SOURCE.read_text(encoding="utf-8")
|
|||
|
|
print(f"知识正文：{SOURCE}（{len(content)} 字）")
|
|||
|
|
|
|||
|
|
with httpx.Client(base_url=args.base, timeout=120) as client:
|
|||
|
|
token = _login(client, args.base, args.username, args.password)
|
|||
|
|
existing = _existing_active(client, args.base, token)
|
|||
|
|
print(f"库内同源未过期知识：{len(existing)} 条 {[r.get('knowledge_id') for r in existing]}")
|
|||
|
|
if existing and not args.force:
|
|||
|
|
print("已存在 → 跳过（要覆盖加 --force）。")
|
|||
|
|
_probe_scores()
|
|||
|
|
return 0
|
|||
|
|
if not args.apply:
|
|||
|
|
print("\n[dry-run] 未上传。加 --apply 真写。")
|
|||
|
|
return 0
|
|||
|
|
|
|||
|
|
for row in existing:
|
|||
|
|
kid = row.get("knowledge_id")
|
|||
|
|
resp = client.delete(f"{args.base}/api/v1/knowledge/{kid}", headers=_headers(token))
|
|||
|
|
print(f" 删除旧条目 {kid}: HTTP {resp.status_code}")
|
|||
|
|
resp = client.post(
|
|||
|
|
f"{args.base}/api/v1/knowledge/upload",
|
|||
|
|
headers=_headers(token),
|
|||
|
|
json={
|
|||
|
|
"filename": FILENAME,
|
|||
|
|
"content_base64": base64.b64encode(content.encode("utf-8")).decode("ascii"),
|
|||
|
|
"knowledge_type": "faq",
|
|||
|
|
},
|
|||
|
|
)
|
|||
|
|
print(f"上传：HTTP {resp.status_code} {resp.text[:200]}")
|
|||
|
|
if resp.status_code != 201:
|
|||
|
|
return 1
|
|||
|
|
print("等 Worker 投向量 …")
|
|||
|
|
time.sleep(12)
|
|||
|
|
_probe_scores()
|
|||
|
|
return 0
|
|||
|
|
|
|||
|
|
|
|||
|
|
if __name__ == "__main__":
|
|||
|
|
raise SystemExit(main())
|