163 lines
5.4 KiB
Python
163 lines
5.4 KiB
Python
|
"""
|
|||
|
|
大模型分析模块 - 调用LLM API分析评论情感
|
|||
|
|
"""
|
|||
|
|
import json
|
|||
|
|
import time
|
|||
|
|
import re
|
|||
|
|
from typing import Dict, Optional, Tuple
|
|||
|
|
from openai import OpenAI, OpenAIError
|
|||
|
|
|
|||
|
|
|
|||
|
|
class LLMAnalyzer:
|
|||
|
|
"""大模型情感分析器"""
|
|||
|
|
|
|||
|
|
SYSTEM_PROMPT = """你是一个专业的情感分析助手。你的任务是分析股吧/论坛评论的情感倾向，判断投资者对该股票的态度。
|
|||
|
|
|
|||
|
|
评分规则：
|
|||
|
|
- 0-30: 极度悲观/看空（利空、暴跌、绝望等情绪）
|
|||
|
|
- 31-50: 偏悲观/中性（担忧、谨慎、观望等情绪）
|
|||
|
|
- 51-70: 偏乐观/中性（看好、希望、期待等情绪）
|
|||
|
|
- 71-100: 极度乐观/看涨（利好、暴涨、兴奋等情绪）
|
|||
|
|
|
|||
|
|
请直接输出一个JSON格式的结果，包含两个字段：
|
|||
|
|
- score: 0-100的整数评分
|
|||
|
|
- label: 简短的态度描述（如"强烈看跌"、"谨慎观望"、"温和看涨"、"强烈看涨"等）
|
|||
|
|
|
|||
|
|
注意：
|
|||
|
|
1. 只返回JSON，不要有其他文字
|
|||
|
|
2. 如果无法判断，返回50和"无法判断"
|
|||
|
|
3. 分析要客观，不要被表面文字迷惑
|
|||
|
|
"""
|
|||
|
|
|
|||
|
|
def __init__(self, config: Dict):
|
|||
|
|
self.config = config
|
|||
|
|
self.base_url = config.get('base_url', 'https://api.openai.com/v1')
|
|||
|
|
self.api_key = config.get('api_key', '')
|
|||
|
|
self.model = config.get('model', 'gpt-3.5-turbo')
|
|||
|
|
self.timeout = config.get('timeout', 30)
|
|||
|
|
self.retry_times = config.get('retry_times', 3)
|
|||
|
|
|
|||
|
|
self.client = None
|
|||
|
|
if self.api_key:
|
|||
|
|
self._init_client()
|
|||
|
|
|
|||
|
|
def _init_client(self):
|
|||
|
|
"""初始化OpenAI客户端"""
|
|||
|
|
try:
|
|||
|
|
self.client = OpenAI(
|
|||
|
|
api_key=self.api_key,
|
|||
|
|
base_url=self.base_url,
|
|||
|
|
timeout=self.timeout
|
|||
|
|
)
|
|||
|
|
except Exception as e:
|
|||
|
|
print(f"初始化LLM客户端失败: {e}")
|
|||
|
|
|
|||
|
|
def update_config(self, config: Dict):
|
|||
|
|
"""更新配置"""
|
|||
|
|
self.config.update(config)
|
|||
|
|
self.base_url = config.get('base_url', self.base_url)
|
|||
|
|
self.api_key = config.get('api_key', self.api_key)
|
|||
|
|
self.model = config.get('model', self.model)
|
|||
|
|
self.timeout = config.get('timeout', self.timeout)
|
|||
|
|
self.retry_times = config.get('retry_times', self.retry_times)
|
|||
|
|
|
|||
|
|
if self.api_key:
|
|||
|
|
self._init_client()
|
|||
|
|
|
|||
|
|
def analyze(self, comment: str) -> Tuple[Optional[int], Optional[str]]:
|
|||
|
|
"""
|
|||
|
|
分析单条评论
|
|||
|
|
返回 (score, label)
|
|||
|
|
"""
|
|||
|
|
if not self.client:
|
|||
|
|
return None, "LLM未配置"
|
|||
|
|
|
|||
|
|
if not comment or not comment.strip():
|
|||
|
|
return None, "评论为空"
|
|||
|
|
|
|||
|
|
for attempt in range(self.retry_times):
|
|||
|
|
try:
|
|||
|
|
response = self.client.chat.completions.create(
|
|||
|
|
model=self.model,
|
|||
|
|
messages=[
|
|||
|
|
{"role": "system", "content": self.SYSTEM_PROMPT},
|
|||
|
|
{"role": "user", "content": f"请分析以下评论的情感倾向：\n\n{comment}"}
|
|||
|
|
],
|
|||
|
|
temperature=0.3,
|
|||
|
|
max_tokens=200
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
result_text = response.choices[0].message.content.strip()
|
|||
|
|
score, label = self._parse_response(result_text)
|
|||
|
|
|
|||
|
|
if score is not None:
|
|||
|
|
return score, label
|
|||
|
|
|
|||
|
|
except OpenAIError as e:
|
|||
|
|
print(f"API调用失败 (尝试 {attempt + 1}/{self.retry_times}): {e}")
|
|||
|
|
if attempt < self.retry_times - 1:
|
|||
|
|
time.sleep(2 ** attempt) # 指数退避
|
|||
|
|
except Exception as e:
|
|||
|
|
print(f"分析过程出错: {e}")
|
|||
|
|
break
|
|||
|
|
|
|||
|
|
return None, "分析失败"
|
|||
|
|
|
|||
|
|
def _parse_response(self, response: str) -> Tuple[Optional[int], Optional[str]]:
|
|||
|
|
"""解析LLM返回的结果"""
|
|||
|
|
try:
|
|||
|
|
# 尝试直接解析JSON
|
|||
|
|
result = json.loads(response)
|
|||
|
|
score = result.get('score', 50)
|
|||
|
|
label = result.get('label', '无法判断')
|
|||
|
|
|
|||
|
|
# 验证分数范围
|
|||
|
|
score = max(0, min(100, int(score)))
|
|||
|
|
|
|||
|
|
return score, label
|
|||
|
|
|
|||
|
|
except json.JSONDecodeError:
|
|||
|
|
# 尝试从文本中提取
|
|||
|
|
pass
|
|||
|
|
|
|||
|
|
# 尝试从文本中提取数字
|
|||
|
|
numbers = re.findall(r'\b(\d{1,3})\b', response)
|
|||
|
|
if numbers:
|
|||
|
|
score = int(numbers[0])
|
|||
|
|
score = max(0, min(100, score))
|
|||
|
|
|
|||
|
|
# 提取标签
|
|||
|
|
label_match = re.search(r'["']([^"']+)["']', response)
|
|||
|
|
if label_match:
|
|||
|
|
label = label_match.group(1)
|
|||
|
|
else:
|
|||
|
|
label = response.split('\n')[0][:20] if response else '无法判断'
|
|||
|
|
|
|||
|
|
return score, label
|
|||
|
|
|
|||
|
|
return None, "解析失败"
|
|||
|
|
|
|||
|
|
def analyze_batch(self, comments: list, delay: float = 1.0) -> list:
|
|||
|
|
"""
|
|||
|
|
批量分析评论
|
|||
|
|
delay: 每次调用之间的延迟（秒）
|
|||
|
|
"""
|
|||
|
|
results = []
|
|||
|
|
|
|||
|
|
for i, comment in enumerate(comments):
|
|||
|
|
print(f"分析评论 {i + 1}/{len(comments)}...")
|
|||
|
|
score, label = self.analyze(comment)
|
|||
|
|
results.append({
|
|||
|
|
'content': comment,
|
|||
|
|
'score': score,
|
|||
|
|
'label': label
|
|||
|
|
})
|
|||
|
|
|
|||
|
|
if delay > 0 and i < len(comments) - 1:
|
|||
|
|
time.sleep(delay)
|
|||
|
|
|
|||
|
|
return results
|
|||
|
|
|
|||
|
|
def is_configured(self) -> bool:
|
|||
|
|
"""检查是否已配置"""
|
|||
|
|
return bool(self.client and self.api_key)
|