168 lines
5.4 KiB
Python
168 lines
5.4 KiB
Python
"""Use DeepSeek to summarize AI/quant research RSS into categorized briefs.
|
||
v2: back-fills pub_time from raw RSS data after AI summarization.
|
||
"""
|
||
import json, os, requests, re
|
||
from datetime import datetime, timezone, timedelta
|
||
from dateutil import parser as dtparser
|
||
|
||
API_KEY = "sk-bbca4a0380d549389f0d27cdea0b5228"
|
||
API_URL = "https://api.deepseek.com/v1/chat/completions"
|
||
MODEL = "deepseek-v4-flash"
|
||
|
||
INPUT = "/var/www/nav/data/research.json"
|
||
OUTPUT = "/var/www/nav/data/research_summary.json"
|
||
tz = timezone(timedelta(hours=8))
|
||
|
||
|
||
def load_raw():
|
||
if not os.path.exists(INPUT):
|
||
return []
|
||
with open(INPUT) as f:
|
||
data = json.load(f)
|
||
return data.get("items", [])
|
||
|
||
|
||
def build_time_lookup(raw_items):
|
||
"""Build lookup: title -> pub_time (Beijing time string)."""
|
||
lookup = {}
|
||
for item in raw_items:
|
||
title = item.get("title", "").strip()
|
||
published = item.get("published", "").strip()
|
||
if not title or not published:
|
||
continue
|
||
try:
|
||
dt = dtparser.parse(published)
|
||
if dt.tzinfo:
|
||
dt = dt.astimezone(tz)
|
||
else:
|
||
dt = dt.replace(tzinfo=tz)
|
||
now = datetime.now(tz)
|
||
if dt.date() == now.date():
|
||
time_str = dt.strftime("%H:%M")
|
||
else:
|
||
time_str = dt.strftime("%m-%d %H:%M")
|
||
lookup[title] = time_str
|
||
except Exception:
|
||
continue
|
||
return lookup
|
||
|
||
|
||
def match_pub_time(items, lookup):
|
||
for item in items:
|
||
if "pub_time" in item:
|
||
continue
|
||
title = item.get("title", "").strip()
|
||
if title in lookup:
|
||
item["pub_time"] = lookup[title]
|
||
else:
|
||
for raw_title, time_val in lookup.items():
|
||
short = min(len(raw_title), len(title))
|
||
if short > 10 and (raw_title[:short] in title or title[:short] in raw_title):
|
||
item["pub_time"] = time_val
|
||
break
|
||
|
||
|
||
def summarize(items):
|
||
news_text = ""
|
||
for i, item in enumerate(items[:100], 1):
|
||
title = item.get("title", "")
|
||
summary = item.get("summary", "")[:300].replace("<", "").replace(">", "")
|
||
source = item.get("source", "")
|
||
link = item.get("link", "")
|
||
news_text += f"{i}. [{source}] {title}\n {summary}\n url: {link}\n\n"
|
||
|
||
prompt = f"""你是一个 AI/量化研究资讯编辑。请从原始 RSS 中精选出**最有价值的论文和博客文章**，按主题分类。
|
||
|
||
## 选稿标准
|
||
- ArXiv 论文：AI 重大突破（LLM、Agent、多模态）、量化交易新方法
|
||
- 博客文章：有实操价值的策略思路、代码实现、回测分析
|
||
- 过滤：纯新闻通告、产品广告、重复性内容
|
||
|
||
## 分类（每个分类 8-15 条）
|
||
- AI 前沿论文（ArXiv cs.AI / cs.LG 重要论文）
|
||
- 量化交易研究（策略思路、回测方法、风险管理）
|
||
- 行业动态（AI 公司、产品发布、监管动态）
|
||
|
||
## 输出格式
|
||
每条：rank, subtitle（15-30字）, title（必须与原始标题完全一致，这是时间匹配的关键）, source, link, summary（200-400字，说清楚这篇讲了什么、有什么用）
|
||
|
||
## 原始内容
|
||
{news_text}
|
||
|
||
JSON格式: {{"categories": [{{"name":"分类名","items":[{{"rank":1,"subtitle":"...","title":"...","source":"...","link":"...","summary":"..."}}]}}]}}"""
|
||
|
||
resp = requests.post(
|
||
API_URL,
|
||
headers={"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"},
|
||
json={
|
||
"model": MODEL,
|
||
"messages": [{"role": "user", "content": prompt}],
|
||
"temperature": 0.3,
|
||
"max_tokens": 16000,
|
||
},
|
||
timeout=180,
|
||
)
|
||
|
||
if resp.status_code != 200:
|
||
print(f"API error: {resp.status_code}")
|
||
return None
|
||
|
||
content = resp.json()["choices"][0]["message"]["content"].strip()
|
||
if content.startswith("```"):
|
||
lines = content.split("\n")
|
||
content = "\n".join(lines[1:-1]) if lines[-1].strip() == "```" else "\n".join(lines[1:])
|
||
|
||
try:
|
||
result = json.loads(content)
|
||
except:
|
||
content = re.sub(r',\s*}', '}', content)
|
||
content = re.sub(r',\s*]', ']', content)
|
||
try:
|
||
result = json.loads(content)
|
||
except:
|
||
print(f"JSON parse failed, len={len(content)}")
|
||
return None
|
||
|
||
return result
|
||
|
||
|
||
def run():
|
||
now = datetime.now(tz).strftime("%Y-%m-%d %H:%M")
|
||
raw_items = load_raw()
|
||
if not raw_items:
|
||
print("No research data")
|
||
exit(0)
|
||
|
||
print(f"Loaded {len(raw_items)} research items. AI summarizing...")
|
||
data = summarize(raw_items)
|
||
|
||
if not data:
|
||
print("Summarization failed")
|
||
exit(1)
|
||
|
||
# Back-fill pub_time
|
||
time_lookup = build_time_lookup(raw_items)
|
||
matched = 0
|
||
for cat in data.get("categories", []):
|
||
match_pub_time(cat["items"], time_lookup)
|
||
matched += sum(1 for item in cat["items"] if "pub_time" in item)
|
||
print(f"Matched pub_time for {matched} articles")
|
||
|
||
result = {
|
||
"updated_at": now,
|
||
"categories": data.get("categories", []),
|
||
}
|
||
|
||
total = sum(len(c["items"]) for c in result["categories"])
|
||
os.makedirs(os.path.dirname(OUTPUT), exist_ok=True)
|
||
with open(OUTPUT, "w", encoding="utf-8") as f:
|
||
json.dump(result, f, ensure_ascii=False, indent=2)
|
||
|
||
print(f"Written {total} research briefs -> {OUTPUT}")
|
||
for c in result["categories"]:
|
||
print(f" {c['name']}: {len(c['items'])} items")
|
||
|
||
|
||
if __name__ == "__main__":
|
||
run()
|