AI回答情感分析
品牌声誉管理实战指南

# AI搜索情感分析 - LLM API标注方案
import json
import requests

def analyze_sentiment_with_llm(text, brand_name, api_key, api_url):
    '''使用LLM API分析品牌提及情感
    
    Args:
        text: AI搜索回答文本
        brand_name: 品牌名称
        api_key: LLM API密钥
        api_url: LLM API地址
    Returns:
        情感分析结果
    '''
    prompt = f'''请分析以下文本中品牌'{brand_name}'的情感倾向。

文本内容:
{text}

请按以下JSON格式返回分析结果:
{{
  'brand_mentioned': true/false,
  'sentiment': 'strongly_positive' / 'positive' / 'neutral' / 'negative' / 'strongly_negative',
  'sentiment_score': 1-5,
  'key_phrases': ['导致此情感判断的关键短语'],
  'competitor_comparison': '如果文本中同时提及竞品，描述品牌vs竞品的对比情感',
  'negative_info_source': '如果是负面提及，说明负面信息来源（如新闻报道/用户投诉/竞品对比）'
}}

注意：
1. 只分析品牌'{brand_name}'的情感，不分析其他品牌
2. 如果品牌未被提及，brand_mentioned设为false，其他字段留空
3. 情感判断应基于品牌提及的上下文（前后100字），而非全文整体语气
'''
    
    headers = {
        'Content-Type': 'application/json',
        'Authorization': f'Bearer {api_key}'
    }
    
    payload = {
        'model': 'gpt-4',
        'messages': [
            {'role': 'user', 'content': prompt}
        ],
        'temperature': 0.1  # 低温度保证结果一致性
    }
    
    response = requests.post(api_url, headers=headers, json=payload)
    result = response.json()
    
    # 解析LLM返回的分析结果
    analysis_text = result['choices'][0]['message']['content']
    try:
        analysis = json.loads(analysis_text)
    except:
        analysis = {'error': 'LLM返回格式错误', 'raw_response': analysis_text}
    
    return analysis

# 批量情感分析
def batch_sentiment_analysis(queries_file, brand_name, api_key, api_url):
    '''批量分析多个AI搜索查询结果的情感'''
    with open(queries_file, 'r', encoding='utf-8') as f:
        queries = json.load(f)
    
    results = []
    for q in queries:
        analysis = analyze_sentiment_with_llm(
            q['ai_response'], brand_name, api_key, api_url
        )
        results.append({
            'date': q['date'],
            'platform': q['platform'],
            'keyword': q['keyword'],
            'brand_mentioned': analysis.get('brand_mentioned', False),
            'sentiment': analysis.get('sentiment', 'unknown'),
            'sentiment_score': analysis.get('sentiment_score', 0),
            'key_phrases': analysis.get('key_phrases', []),
            'negative_source': analysis.get('negative_info_source', '')
        })
    
    return results

# 情感汇总统计
def summarize_sentiment(results):
    '''汇总情感分析结果'''
    from collections import Counter
    mentioned = [r for r in results if r['brand_mentioned']]
    sentiment_dist = Counter(r['sentiment'] for r in mentioned)
    total = len(mentioned)
    
    print(f'品牌提及总数: {total}/{len(results)}')
    print(f'提及率: {total/len(results)*100:.1f}%')
    print(f'\n情感分布:')
    for sent in ['strongly_positive', 'positive', 'neutral', 'negative', 'strongly_negative']:
        count = sentiment_dist.get(sent, 0)
        pct = count / total * 100 if total > 0 else 0
        print(f'  {sent}: {count} ({pct:.1f}%)')
    
    positive_pct = (sentiment_dist.get('strongly_positive', 0) + sentiment_dist.get('positive', 0)) / total * 100 if total else 0
    negative_pct = (sentiment_dist.get('negative', 0) + sentiment_dist.get('strongly_negative', 0)) / total * 100 if total else 0
    print(f'\n正面提及比例: {positive_pct:.1f}%')
    print(f'负面提及比例: {negative_pct:.1f}%')
    print(f'品牌净情感分: {positive_pct - negative_pct:.1f}')

# 负面信息预警系统

import smtplib
from email.mime.text import MIMEText
from datetime import datetime, timedelta

class SentimentAlertSystem:
    def __init__(self, brand_name, alert_email):
        self.brand_name = brand_name
        self.alert_email = alert_email
        self.alert_rules = [
            {
                'name': '负面比例突增',
                'condition': lambda current, history: current > history * 2,
                'severity': 'high',
                'message': '负面提及比例超过历史平均值的2倍'
            },
            {
                'name': '强烈负面首次出现',
                'condition': lambda current, history: current > 0 and history == 0,
                'severity': 'critical',
                'message': 'AI回答中首次出现强烈负面品牌提及'
            },
            {
                'name': '特定平台连续负面',
                'condition': lambda current, history: current >= 3,
                'severity': 'high',
                'message': '某平台连续3次查询出现负面提及'
            },
            {
                'name': '竞品对比负面',
                'condition': lambda current, history: current > 0,
                'severity': 'medium',
                'message': '竞品对比查询中出现对本品的负面描述'
            }
        ]
    
    def check_alerts(self, current_data, historical_data):
        '''检查预警规则'''
        alerts = []
        for rule in self.alert_rules:
            # current_data格式: {'negative_pct': 16, 'strongly_negative_count': 1, ...}
            # 根据规则类型获取对应数据
            if rule['name'] == '负面比例突增':
                triggered = rule['condition'](
                    current_data.get('negative_pct', 0),
                    historical_data.get('avg_negative_pct', 0)
                )
            elif rule['name'] == '强烈负面首次出现':
                triggered = rule['condition'](
                    current_data.get('strongly_negative_count', 0),
                    historical_data.get('prev_strongly_negative_count', 0)
                )
            elif rule['name'] == '特定平台连续负面':
                triggered = rule['condition'](
                    current_data.get('consecutive_negative', 0), 0
                )
            elif rule['name'] == '竞品对比负面':
                triggered = rule['condition'](
                    current_data.get('comparison_negative', 0), 0
                )
            
            if triggered:
                alerts.append({
                    'rule_name': rule['name'],
                    'severity': rule['severity'],
                    'message': rule['message'],
                    'current_data': current_data,
                    'timestamp': datetime.now().isoformat()
                })
        return alerts
    
    def send_alert(self, alerts):
        '''发送告警邮件'''
        if not alerts:
            return
        
        subject = f'[GEO预警] {self.brand_name} 品牌声誉风险 ({len(alerts)}条告警)'
        body = f'品牌: {self.brand_name}\n时间: {datetime.now().strftime('%Y-%m-%d %H:%M')}\n\n'
        for alert in alerts:
            body += f'告警: {alert['rule_name']}\n'
            body += f'严重程度: {alert['severity']}\n'
            body += f'描述: {alert['message']}\n'
            body += f'当前数据: {alert['current_data']}\n'
            body += '-' * 40 + '\n'
        
        body += '\n请GEO负责人在1小时内确认并在24小时内提交应对方案。'
        
        print(f'告警邮件已发送: {subject}')
        print(f'收件人: {self.alert_email}')
        print(f'告警数量: {len(alerts)}')
        for a in alerts:
            print(f'  [{a['severity']}] {a['rule_name']}: {a['message']}')

# 使用示例
alert_system = SentimentAlertSystem('SmartCRM', 'geo-alert@company.com')
current = {'negative_pct': 16, 'strongly_negative_count': 1, 'consecutive_negative': 0, 'comparison_negative': 0}
historical = {'avg_negative_pct': 6, 'prev_strongly_negative_count': 0}

alerts = alert_system.check_alerts(current, historical)
if alerts:
    alert_system.send_alert(alerts)