125 lines
5.6 KiB
Python
125 lines
5.6 KiB
Python
# -*- coding: utf-8 -*-
|
||
"""
|
||
阿里云 DashScope（通义千问）提取面板结构化数据
|
||
从新闻文本中提取战损、报复指数、基地状态等，供 db_merge 落库
|
||
API Key 通过环境变量 DASHSCOPE_API_KEY 配置
|
||
"""
|
||
import json
|
||
import os
|
||
import re
|
||
from datetime import datetime, timezone
|
||
from typing import Any, Dict, Optional
|
||
|
||
from panel_schema import validate_category, validate_severity, validate_summary
|
||
|
||
|
||
def _call_dashscope_extract(text: str, timeout: int = 15) -> Optional[Dict[str, Any]]:
|
||
"""调用阿里云 DashScope 提取结构化数据"""
|
||
api_key = os.environ.get("DASHSCOPE_API_KEY", "").strip()
|
||
if not api_key or not text or len(str(text).strip()) < 10:
|
||
return None
|
||
try:
|
||
import dashscope
|
||
from http import HTTPStatus
|
||
|
||
dashscope.api_key = api_key
|
||
|
||
prompt = f"""从以下美伊/中东军事新闻中提取可明确推断的数值，输出 JSON。无依据的字段省略不写。
|
||
|
||
要求：
|
||
- summary: 1-2句中文事实摘要，≤80字
|
||
- category: deployment|alert|intel|diplomatic|other
|
||
- severity: low|medium|high|critical
|
||
- 战损（仅当新闻明确提及数字时填写）:
|
||
us_personnel_killed, iran_personnel_killed, us_personnel_wounded, iran_personnel_wounded,
|
||
us_civilian_killed, iran_civilian_killed, us_civilian_wounded, iran_civilian_wounded,
|
||
us_bases_destroyed, iran_bases_destroyed, us_bases_damaged, iran_bases_damaged.
|
||
重要：bases_* 仅指已确认损毁/受损的基地数量；"军事目标"/"targets"等泛指不是基地，若报道只说"X个军事目标遭袭"而无具体基地名，不填写 bases_*
|
||
us_aircraft, iran_aircraft, us_warships, iran_warships, us_armor, iran_armor, us_vehicles, iran_vehicles,
|
||
us_drones, iran_drones, us_missiles, iran_missiles, us_helicopters, iran_helicopters, us_submarines, iran_submarines
|
||
- retaliation_sentiment: 0-100，仅当新闻涉及伊朗报复/反击情绪时
|
||
- wall_street_value: 0-100，仅当新闻涉及美股/市场反应时
|
||
- key_location_updates: 当新闻提及具体基地/设施遭袭时必填，数组 [{{"name_keywords":"阿萨德|asad|assad|阿因","side":"us","status":"attacked","damage_level":1-3}}]。常用关键词：阿萨德|asad|巴格达|baghdad|乌代德|udeid|埃尔比勒|erbil|因吉尔利克|incirlik|德黑兰|tehran|阿巴斯|abbas|布什尔|bushehr|伊斯法罕|isfahan|纳坦兹|natanz
|
||
|
||
原文：
|
||
{str(text)[:800]}
|
||
|
||
直接输出 JSON，不要其他解释："""
|
||
|
||
response = dashscope.Generation.call(
|
||
model="qwen-turbo",
|
||
messages=[{"role": "user", "content": prompt}],
|
||
result_format="message",
|
||
max_tokens=512,
|
||
)
|
||
|
||
if response.status_code != HTTPStatus.OK:
|
||
return None
|
||
raw = (response.output.get("choices", [{}])[0].get("message", {}).get("content", "") or "").strip()
|
||
raw = re.sub(r"^```\w*\s*|\s*```$", "", raw)
|
||
return json.loads(raw)
|
||
except Exception:
|
||
return None
|
||
|
||
|
||
def extract_from_news(text: str, timestamp: Optional[str] = None) -> Dict[str, Any]:
|
||
"""
|
||
从新闻文本提取结构化数据，符合面板 schema
|
||
返回: { situation_update?, combat_losses_delta?, retaliation?, wall_street?, key_location_updates? }
|
||
"""
|
||
ts = timestamp or datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%S.000Z")
|
||
out: Dict[str, Any] = {}
|
||
parsed = _call_dashscope_extract(text)
|
||
if not parsed:
|
||
return out
|
||
|
||
if parsed.get("summary"):
|
||
out["situation_update"] = {
|
||
"summary": validate_summary(str(parsed["summary"])[:120], 120),
|
||
"category": validate_category(str(parsed.get("category", "other")).lower()),
|
||
"severity": validate_severity(str(parsed.get("severity", "medium")).lower()),
|
||
"timestamp": ts,
|
||
}
|
||
|
||
loss_us = {}
|
||
loss_ir = {}
|
||
for k in ["personnel_killed", "personnel_wounded", "civilian_killed", "civilian_wounded",
|
||
"bases_destroyed", "bases_damaged", "aircraft", "warships", "armor", "vehicles",
|
||
"drones", "missiles", "helicopters", "submarines"]:
|
||
uk, ik = f"us_{k}", f"iran_{k}"
|
||
if uk in parsed and isinstance(parsed[uk], (int, float)):
|
||
loss_us[k] = max(0, int(parsed[uk]))
|
||
if ik in parsed and isinstance(parsed[ik], (int, float)):
|
||
loss_ir[k] = max(0, int(parsed[ik]))
|
||
if loss_us or loss_ir:
|
||
out["combat_losses_delta"] = {}
|
||
if loss_us:
|
||
out["combat_losses_delta"]["us"] = loss_us
|
||
if loss_ir:
|
||
out["combat_losses_delta"]["iran"] = loss_ir
|
||
|
||
if "retaliation_sentiment" in parsed:
|
||
v = parsed["retaliation_sentiment"]
|
||
if isinstance(v, (int, float)) and 0 <= v <= 100:
|
||
out["retaliation"] = {"value": int(v), "time": ts}
|
||
|
||
if "wall_street_value" in parsed:
|
||
v = parsed["wall_street_value"]
|
||
if isinstance(v, (int, float)) and 0 <= v <= 100:
|
||
out["wall_street"] = {"time": ts, "value": int(v)}
|
||
|
||
if "key_location_updates" in parsed and isinstance(parsed["key_location_updates"], list):
|
||
valid = []
|
||
for u in parsed["key_location_updates"]:
|
||
if isinstance(u, dict) and u.get("name_keywords") and u.get("side") in ("us", "iran"):
|
||
valid.append({
|
||
"name_keywords": str(u["name_keywords"]),
|
||
"side": u["side"],
|
||
"status": str(u.get("status", "attacked"))[:20],
|
||
"damage_level": min(3, max(1, int(u["damage_level"]))) if isinstance(u.get("damage_level"), (int, float)) else 2,
|
||
})
|
||
if valid:
|
||
out["key_location_updates"] = valid
|
||
|
||
return out
|