Python批量分析
AI搜索回答深度解析方法

# 数据预处理流程
import pandas as pd
import re
import json

# 加载JSONL格式的检测结果
def load_results(filepath):
    records = []
    with open(filepath, 'r', encoding='utf-8') as f:
        for line in f:
            records.append(json.loads(line.strip()))
    return pd.DataFrame(records)

# 文本清洗
def clean_text(text):
    if not isinstance(text, str):
        return ''
    # 去除HTML标签
    text = re.sub(r'<[^>]+>', '', text)
    # 去除特殊字符
    text = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f]', '', text)
    # 统一空白字符
    text = re.sub(r'\s+', ' ', text).strip()
    return text

# 数据预处理管道
def preprocess_data(df):
    # 文本清洗
    df['response_clean'] = df['response'].apply(clean_text)
    # 去除空回答
    df = df[df['response_clean'].str.len() > 50].copy()
    # 去重（同关键词同平台同日保留最新）
    df['date'] = pd.to_datetime(df['timestamp'], unit='ms').dt.date
    df = df.sort_values('timestamp').drop_duplicates(
        subset=['keyword', 'platform', 'date'], keep='last'
    )
    # 重置索引
    df = df.reset_index(drop=True)
    print(f'预处理完成: {len(df)} 条有效记录')
    return df

# 使用示例
# df = load_results('results/detection_20260811.jsonl')
# df = preprocess_data(df)
# print(df[['platform', 'keyword', 'brand_mentioned']].head(10))

# 品牌提及检测与位置分析
def detect_mention(text, brand_names):
    '''多层品牌提及检测'''
    results = []
    text_lower = text.lower()
    for name in brand_names:
        positions = []
        start = 0
        while True:
            pos = text_lower.find(name.lower(), start)
            if pos == -1:
                break
            positions.append(pos)
            start = pos + len(name)
        if positions:
            results.append({
                'brand': name,
                'count': len(positions),
                'first_position': positions[0],
                'relative_position': positions[0] / len(text) if len(text) > 0 else 1.0,
                'all_positions': positions
            })
    return results

def classify_position(relative_pos):
    '''将相对位置分类'''
    if relative_pos < 0.33:
        return '前1/3 (高优先级)'
    elif relative_pos < 0.66:
        return '中1/3 (中等优先级)'
    else:
        return '后1/3 (低优先级)'

# 批量分析
brand_variants = ['你的品牌名', '品牌简称', 'BrandName']

# for idx, row in df.iterrows():
#     mentions = detect_mention(row['response_clean'], brand_variants)
#     if mentions:
#         pos_label = classify_position(mentions[0]['relative_position'])
#         print(f'{row["platform"]} | {row["keyword"]} | 提及{mentions[0]["count"]}次 | 位置: {pos_label}')

# 使用LLM API进行情感分析
import json
import requests

def analyze_sentiment(text, brand_name, api_key):
    '''使用LLM分析AI回答中品牌的情感倾向'''
    prompt = f'''请分析以下AI搜索回答中关于品牌"{brand_name}"的情感倾向。

AI回答内容：
{text}

请以JSON格式返回分析结果，包含以下字段：
- sentiment: 情感倾向 (positive/negative/neutral)
- intensity: 情感强度 (strong/medium/weak)
- key_phrases: 关键评价语句列表
- competitor_comparison: 与竞品的对比描述（如果有）
- overall_recommendation: AI是否推荐该品牌 (yes/no/partial)
'''

    response = requests.post(
        'https://api.openai.com/v1/chat/completions',
        headers={'Authorization': f'Bearer {api_key}'},
        json={
            'model': 'gpt-4o-mini',
            'messages': [{'role': 'user', 'content': prompt}],
            'temperature': 0.3
        }
    )

    result = response.json()['choices'][0]['message']['content']
    try:
        return json.loads(result)
    except json.JSONDecodeError:
        return {'sentiment': 'unknown', 'raw': result}

# 批量情感分析
def batch_sentiment_analysis(df, brand_name, api_key):
    sentiments = []
    for idx, row in df.iterrows():
        if row.get('brand_mentioned'):
            analysis = analyze_sentiment(row['response_clean'], brand_name, api_key)
            sentiments.append(analysis)
            print(f'{row["platform"]} | {row["keyword"]} | {analysis.get("sentiment", "?")}')
        else:
            sentiments.append({'sentiment': 'not_mentioned'})
    df['sentiment'] = [s.get('sentiment', 'unknown') for s in sentiments]
    return df

# 主题聚类分析
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.cluster import KMeans
import numpy as np

def cluster_responses(texts, n_clusters=8):
    '''对AI回答文本进行主题聚类'''
    # TF-IDF向量化
    vectorizer = TfidfVectorizer(max_features=1000, stop_words='english')
    tfidf_matrix = vectorizer.fit_transform(texts)

    # K-Means聚类
    kmeans = KMeans(n_clusters=n_clusters, random_state=42, n_init=10)
    clusters = kmeans.fit_predict(tfidf_matrix)

    # 提取每个聚类的关键词
    feature_names = vectorizer.get_feature_names_out()
    cluster_keywords = {}
    for i in range(n_clusters):
        center = kmeans.cluster_centers_[i]
        top_indices = center.argsort()[-10:][::-1]
        keywords = [feature_names[idx] for idx in top_indices]
        cluster_keywords[i] = keywords
        print(f'Cluster {i}: {", ".join(keywords[:5])}')

    return clusters, cluster_keywords

def analyze_cluster_mention_rate(df, clusters, brand_name):
    '''分析各聚类的品牌提及率'''
    df['cluster'] = clusters
    for cluster_id in df['cluster'].unique():
        subset = df[df['cluster'] == cluster_id]
        mention_rate = subset['brand_mentioned'].mean()
        count = len(subset)
        print(f'Cluster {cluster_id}: {count}条 | 提及率 {mention_rate:.1%}')

# 可视化报告生成
import matplotlib.pyplot as plt
import matplotlib
matplotlib.rcParams['font.sans-serif'] = ['SimHei']
matplotlib.rcParams['axes.unicode_minus'] = False

def generate_report_charts(df, brand_name):
    '''生成GEO效果分析图表'''
    fig, axes = plt.subplots(2, 2, figsize=(14, 10))

    # 1. 品牌提及率趋势
    daily = df.groupby('date')['brand_mentioned'].mean()
    axes[0][0].plot(daily.index, daily.values, marker='o')
    axes[0][0].set_title('品牌提及率趋势')
    axes[0][0].set_ylabel('提及率')

    # 2. 平台对比
    platform = df.groupby('platform')['brand_mentioned'].mean()
    axes[0][1].bar(platform.index, platform.values)
    axes[0][1].set_title('各平台提及率对比')
    axes[0][1].set_ylabel('提及率')

    # 3. 情感分布
    sentiment = df[df['brand_mentioned']]['sentiment'].value_counts()
    axes[1][0].pie(sentiment.values, labels=sentiment.index, autopct='%1.1f%%')
    axes[1][0].set_title('品牌提及情感分布')

    # 4. 提及位置分布
    axes[1][1].set_title('分析报告完成')
    axes[1][1].text(0.5, 0.5, f'品牌: {brand_name}\n检测: {len(df)}条\n提及: {df["brand_mentioned"].sum()}条',
                    ha='center', va='center', fontsize=14)
    axes[1][1].axis('off')

    plt.tight_layout()
    plt.savefig('geo_report.png', dpi=150, bbox_inches='tight')
    print('报告已保存: geo_report.png')

# generate_report_charts(df, '你的品牌名')