143 lines
5.3 KiB
Python
143 lines
5.3 KiB
Python
"""
|
||
LLM服务
|
||
负责：调用Agent进行舆情分析
|
||
"""
|
||
|
||
import json
|
||
import asyncio
|
||
from datetime import datetime
|
||
from typing import Dict
|
||
|
||
class LLMService:
|
||
"""LLM舆情分析服务"""
|
||
|
||
def __init__(self):
|
||
# 模拟新闻数据源
|
||
self.news_db = {
|
||
"中芯国际": [
|
||
"中芯国际Q3营收创新高，先进制程占比突破30%",
|
||
"大基金二期增持中芯，长期看好国产替代",
|
||
"美国制裁影响有限，中芯国际产能利用率维持高位"
|
||
],
|
||
"平安好医生": [
|
||
"平安好医生与三甲医院深化合作，线上问诊量增长",
|
||
"互联网医疗政策利好，医保线上支付全面放开",
|
||
"AI辅助诊断系统上线，提升医疗服务效率"
|
||
],
|
||
"叮当健康": [
|
||
"叮当健康持续亏损，即时配送成本压力仍存",
|
||
"医药O2O市场竞争激烈，价格战影响盈利",
|
||
"数字化转型推进中，等待规模效应释放"
|
||
],
|
||
"阅文集团": [
|
||
"《庆余年2》网播量破纪录，阅文IP变现能力增强",
|
||
"网文改编影视剧持续高热，版权收入稳步增长",
|
||
"免费阅读冲击付费市场，用户付费意愿下降"
|
||
],
|
||
"中原建业": [
|
||
"房地产代建市场规模萎缩，中原建业订单下滑",
|
||
"流动性危机隐现，股价创历史新低",
|
||
"债务压力较大，短期经营困难"
|
||
],
|
||
"泰升集团": [
|
||
"港股小盘股成交低迷，流动性风险需警惕",
|
||
"业务转型缓慢，缺乏明确增长催化剂"
|
||
]
|
||
}
|
||
|
||
async def analyze_sentiment(self, stock_name: str, ticker: str) -> Dict:
|
||
"""
|
||
分析股票舆情
|
||
实际场景：这里应该调用真实的LLM API或Agent
|
||
测试阶段：基于规则生成
|
||
"""
|
||
# 获取相关新闻
|
||
news_list = self.news_db.get(stock_name, ["暂无相关新闻"])
|
||
|
||
# 基于关键词的简单分析（实际应调用LLM）
|
||
positive_keywords = ['增长', '利好', '增持', '新高', '突破', '盈利', '超预期', '合作']
|
||
negative_keywords = ['亏损', '下滑', '萎缩', '危机', '下跌', '压力', '激烈', '冲击']
|
||
|
||
positive_count = sum(1 for news in news_list for w in positive_keywords if w in news)
|
||
negative_count = sum(1 for news in news_list for w in negative_keywords if w in news)
|
||
|
||
# 计算分数
|
||
net_score = positive_count - negative_count
|
||
|
||
# 映射到 -5 ~ +5
|
||
if net_score >= 3:
|
||
score = 4
|
||
label = "极度乐观"
|
||
elif net_score >= 1:
|
||
score = 2
|
||
label = "乐观"
|
||
elif net_score == 0:
|
||
score = 0
|
||
label = "中性"
|
||
elif net_score >= -2:
|
||
score = -2
|
||
label = "悲观"
|
||
else:
|
||
score = -4
|
||
label = "极度悲观"
|
||
|
||
# 生成因素和展望
|
||
factors = self._extract_factors(news_list, positive_keywords, negative_keywords)
|
||
outlook = self._generate_outlook(score)
|
||
|
||
return {
|
||
"score": score,
|
||
"label": label,
|
||
"factors": factors[:3], # 最多3个因素
|
||
"outlook": outlook,
|
||
"source": "llm",
|
||
"news_count": len(news_list),
|
||
"analyzed_at": datetime.now().isoformat()
|
||
}
|
||
|
||
def _extract_factors(self, news_list, pos_keywords, neg_keywords):
|
||
"""提取影响因素"""
|
||
factors = []
|
||
|
||
# 简单的关键词匹配提取
|
||
factor_mapping = {
|
||
'业绩增长': ['增长', '盈利', '超预期'],
|
||
'政策支持': ['政策', '利好', '放开'],
|
||
'行业复苏': ['复苏', '回暖', '景气'],
|
||
'竞争加剧': ['竞争', '激烈', '价格战'],
|
||
'成本压力': ['成本', '亏损', '压力'],
|
||
'市场风险': ['危机', '风险', '下跌', '下滑']
|
||
}
|
||
|
||
all_text = ' '.join(news_list)
|
||
|
||
for factor, keywords in factor_mapping.items():
|
||
if any(kw in all_text for kw in keywords):
|
||
factors.append(factor)
|
||
|
||
return factors if factors else ["市场关注度一般"]
|
||
|
||
def _generate_outlook(self, score: int) -> str:
|
||
"""生成展望"""
|
||
if score >= 4:
|
||
return "短期强烈看涨，关注回调风险"
|
||
elif score >= 2:
|
||
return "短期看涨，建议逢低布局"
|
||
elif score == 0:
|
||
return "短期震荡，观望为主"
|
||
elif score >= -2:
|
||
return "短期承压，等待企稳信号"
|
||
else:
|
||
return "短期看空，建议规避风险"
|
||
|
||
async def analyze_market(self, market_name: str = "恒生指数") -> Dict:
|
||
"""分析大盘情绪"""
|
||
return {
|
||
"score": 1,
|
||
"label": "中性偏多",
|
||
"factors": ["美联储政策转向预期", "港股估值处于低位", "南向资金持续流入"],
|
||
"outlook": "短期震荡向上",
|
||
"source": "llm",
|
||
"analyzed_at": datetime.now().isoformat()
|
||
}
|