News Sentiment Finbert

Huntly (RSS 抓取, 381 源, 41.5 万文章)
  → celery-beat 每 60s: news-enrich-recent (run_enrichment_batch)
  → NewsMatcher (Aho-Corasick + finance_lexicon 5 万词: 股票/行业/事件/情感)
  → sentiment.score() (两级融合)
      字典法 dict_score (finance_lexicon sentiment_pos/neg 词权重差)
      + FinBERT (bardsai/finance-sentiment-zh-base, CPU, conf≥0.55 才融合)
  → 写入 news_article_enrichment 表 (model_version 带 +finbert 后缀表示 FinBERT 生效)
  → 前端:
      /rss-news  NewsPanel: 情绪筛选/利好利空强度排序/统计栏情绪分布条形图
      个股终端「个股资讯」Tab: 每篇利好/利空标签 (stock_news 关联 enrichment)

# 词表已固化: backend/scripts/data/finance_sentiment_lexicon.tsv (51,213 词)
# 导入脚本: backend/scripts/import_sentiment_lexicon.py
# 本地
python3 backend/scripts/import_sentiment_lexicon.py
# 容器
docker exec quantmind python3 /app/backend/scripts/import_sentiment_lexicon.py

curl -s -X POST -H "$AUTH" "$BASE/api/v1/news/enrichment/run"  # 触发重算

docker exec quantmind python3 -c "
import asyncio
from backend.shared.database_manager_v2 import get_session
from sqlalchemy import text
async def main():
    async with get_session() as s:
        r = await s.execute(text(\"SELECT sentiment_label, COUNT(*) FROM news_article_enrichment WHERE model_version LIKE '%finbert%' GROUP BY sentiment_label ORDER BY 2 DESC\"))
        print('+finbert 分布:', r.fetchall())
        r = await s.execute(text(\"SELECT COUNT(*) FROM news_article_enrichment WHERE model_version NOT LIKE '%finbert%'\"))
        print('无 finbert 条数:', r.fetchone()[0])
asyncio.run(main())
"

docker exec quantmind python3 -c "import transformers; print(transformers.__version__)"
# celery worker 也要有（enrich/rebuild 在 celery 里跑）
docker exec quantmind-celery python3 -c "import transformers; print(transformers.__version__)"

docker exec quantmind pip install --no-cache-dir transformers
docker exec quantmind-celery pip install --no-cache-dir transformers

# 触发下载（首次）
docker exec quantmind python3 -c "
from backend.services.api.news import sentiment as s
s._model_ready=False; s._model_failed=False
s._try_load(); print('ready=', s._model_ready)
"
# 权重缓存位置: /root/.cache/huggingface/hub/models--bardsai--finance-sentiment-zh-base
# celery worker 复用（主容器导出 → tar 管道 → celery 导入）
docker exec quantmind tar -C /root/.cache/huggingface/hub -cf - models--bardsai--finance-sentiment-zh-base \
  | docker exec -i quantmind-celery sh -c 'mkdir -p /root/.cache/huggingface/hub && tar -C /root/.cache/huggingface/hub -xf -'

docker exec quantmind-celery python3 -c "
import os
os.environ['HF_HUB_OFFLINE']='1'
from transformers import pipeline
p = pipeline('sentiment-analysis', model='bardsai/finance-sentiment-zh-base', device=-1)
print(p('公司业绩暴雷，股价暴跌')[0])   # 期望 label=negative, score>0.9
"

docker restart quantmind-celery
sleep 8
# 确认后台加载成功
docker exec quantmind-celery python3 -c "
from backend.services.api.news import sentiment as s
s._ensure_loading()
import time
for _ in range(40):
    if s.is_available(): break
    time.sleep(1)
print('FinBERT ready:', s.is_available())
"

TOKEN=$(curl -s -X POST $BASE/api/v1/auth/login -H "Content-Type: application/json" \
  -d '{"username":"admin","password":"admin123","tenant_id":"default"}' \
  | python3 -c "import sys,json; print(json.load(sys.stdin).get('access_token',''))")

# 全量重算（force=true 覆盖已 enrich 的）
curl -s -X POST -H "Authorization: Bearer $TOKEN" \
  "$BASE/api/v1/news/enrichment/rebuild-all?force=true"
# 或增量（跳过 model_version 已是最新的）
curl -s -X POST -H "Authorization: Bearer $TOKEN" \
  "$BASE/api/v1/news/enrichment/rebuild-all"

# 查进度
curl -s -H "Authorization: Bearer $TOKEN" "$BASE/api/v1/news/enrichment/rebuild-progress"

docker exec quantmind python3 /app/backend/scripts/seed_a_share_stocks.py

docker exec quantmind python3 -c "
import asyncio
from backend.shared.database_manager_v2 import get_session
from sqlalchemy import text
async def main():
    async with get_session() as s:
        await s.execute(text(\"INSERT INTO finance_lexicon (term, kind, weight, enabled) VALUES ('涨停','sentiment_pos',1.0,true)\"))
        await s.commit()
asyncio.run(main())
"