mirror of
https://github.com/stardrophere/InsightRadar.git
synced 2026-06-06 00:57:51 +08:00
stardrophere
105 lines
3.8 KiB
Python
105 lines
3.8 KiB
Python
# app/services/summary_service.py
|
|
import os
|
|
import json
|
|
from datetime import timedelta
|
|
from openai import AsyncOpenAI
|
|
|
|
from app.database import SessionLocal
|
|
from app.models.models import UnifiedEvent, TrendingEvent, InfoSource, utcnow
|
|
from app.prompts.summary_prompts import (
|
|
SUMMARY_SYSTEM_PROMPT,
|
|
SUMMARY_USER_PROMPT_TEMPLATE,
|
|
)
|
|
|
|
HOT_SCORE_THRESHOLD = int(os.getenv("HOT_SCORE_THRESHOLD", 3))
|
|
AI_API_KEY = os.getenv("AI_API_KEY", '')
|
|
|
|
# 1. 初始化异步客户端 (全局复用)
|
|
deepseek_client = AsyncOpenAI(
|
|
api_key=AI_API_KEY,
|
|
base_url="https://api.deepseek.com"
|
|
)
|
|
|
|
|
|
async def call_llm_for_summary(platform_data_text: str) -> dict:
|
|
"""调用 DeepSeek 生成统一标题和多平台视角摘要"""
|
|
prompt = SUMMARY_USER_PROMPT_TEMPLATE.format(
|
|
platform_data_text=platform_data_text
|
|
)
|
|
|
|
# await
|
|
response = await deepseek_client.chat.completions.create(
|
|
model="deepseek-chat",
|
|
messages=[
|
|
{"role": "system", "content": SUMMARY_SYSTEM_PROMPT},
|
|
{"role": "user", "content": prompt}
|
|
],
|
|
response_format={"type": "json_object"},
|
|
temperature=1
|
|
)
|
|
|
|
result_text = response.choices[0].message.content
|
|
return json.loads(result_text)
|
|
|
|
|
|
async def generate_unified_summaries():
|
|
"""定时任务：扫描高热度事件并生成/更新摘要"""
|
|
print(f"[{utcnow()}] 开始执行 DeepSeek 摘要生成任务...")
|
|
|
|
with SessionLocal() as db:
|
|
recent_threshold = utcnow() - timedelta(days=3)
|
|
|
|
# 必须满足：热度达标 AND (当前热度 > 上次生成摘要时的热度) AND 近期活跃
|
|
events = db.query(UnifiedEvent).filter(
|
|
UnifiedEvent.hot_score >= HOT_SCORE_THRESHOLD,
|
|
UnifiedEvent.hot_score > UnifiedEvent.last_summarized_trends_count,
|
|
UnifiedEvent.created_at >= recent_threshold
|
|
).all()
|
|
|
|
if not events:
|
|
print("当前没有需要更新摘要的大事件，任务结束。")
|
|
return
|
|
|
|
for event in events:
|
|
# 联合查询获取该事件在各平台的子新闻
|
|
trends = db.query(TrendingEvent, InfoSource.source_name) \
|
|
.join(InfoSource, TrendingEvent.source_id == InfoSource.id) \
|
|
.filter(TrendingEvent.unified_event_id == event.id) \
|
|
.all()
|
|
|
|
if not trends:
|
|
continue
|
|
|
|
# 按平台归类标题并去重
|
|
platform_dict = {}
|
|
for trend_record, source_name in trends:
|
|
if source_name not in platform_dict:
|
|
platform_dict[source_name] = set()
|
|
platform_dict[source_name].add(trend_record.current_headline)
|
|
|
|
# 组装给大模型的 Prompt 数据
|
|
prompt_lines = [f"【{platform}】: {', '.join(headlines)}" for platform, headlines in platform_dict.items()]
|
|
platform_data_text = "\n".join(prompt_lines)
|
|
|
|
try:
|
|
# 调用封装好的异步函数
|
|
llm_result = await call_llm_for_summary(platform_data_text)
|
|
|
|
if "unified_title" in llm_result:
|
|
event.unified_title = llm_result["unified_title"]
|
|
if "ai_comprehensive_summary" in llm_result:
|
|
event.ai_comprehensive_summary = llm_result["ai_comprehensive_summary"]
|
|
|
|
# 成功后更新水位线
|
|
# 将最后一次总结时的热搜数量，更新为当前最新的 hot_score
|
|
event.last_summarized_trends_count = event.hot_score
|
|
|
|
print(f"成功更新大事件 ID {event.id} 的深度摘要 (当前热度: {event.hot_score})。")
|
|
|
|
except Exception as e:
|
|
print(f"大事件 ID {event.id} 摘要生成失败: {e}")
|
|
continue
|
|
|
|
# 提交事务
|
|
db.commit()
|