智能内容审核：守护平台安全

class ContentModerator:
    """智能内容审核系统"""
    
    def __init__(self, llm_api):
        self.llm = llm_api
        self.categories = {
            'violence': '暴力血腥内容',
            'adult': '色情低俗内容',
            'illegal': '违法违规信息',
            'hate': '仇恨歧视言论',
            'privacy': '个人隐私信息',
            'spam': '垃圾广告信息'
        }
    
    def moderate(self, content):
        """审核内容"""
        prompt = f"""请审核以下内容，判断是否包含违规信息。

内容：{content}

请分析以下几个方面：
1. 是否包含暴力、色情、违法等违规内容
2. 是否包含个人隐私或敏感信息
3. 是否为垃圾广告或无意义内容
4. 整体内容质量评分（1-10分）

返回JSON格式：
{{
    "safe": true/false,
    "categories": ["违规类别"],
    "severity": "low/medium/high",
    "reasons": ["具体原因"],
    "score": 1-10,
    "suggestion": "处理建议"
}}"""
        
        response = self.llm.generate(prompt, temperature=0.1)
        return json.loads(response)
    
    def batch_moderate(self, contents, parallel=True):
        """批量审核"""
        if parallel:
            # 并行处理
            with ThreadPoolExecutor(max_workers=10) as executor:
                futures = [
                    executor.submit(self.moderate, content) 
                    for content in contents
                ]
                results = [f.result() for f in futures]
        else:
            # 串行处理
            results = [self.moderate(content) for content in contents]
        
        return results
    
    def real_time_filter(self, text_stream):
        """实时流式审核"""
        buffer = ""
        for chunk in text_stream:
            buffer += chunk
            
            # 每50个字符检查一次
            if len(buffer) > 50:
                result = self.quick_check(buffer)
                if not result['safe']:
                    # 立即中断
                    return {
                        'blocked': True,
                        'reason': result['reason']
                    }
                buffer = buffer[-25:]  # 保留部分用于上下文
        
        # 最终完整检查
        return self.moderate(text_stream.get_full_text())
    
    def custom_rules(self, content, rules):
        """自定义规则审核"""
        violations = []
        
        for rule in rules:
            if rule['type'] == 'keyword':
                if rule['pattern'] in content.lower():
                    violations.append(rule['action'])
            elif rule['type'] == 'regex':
                if re.search(rule['pattern'], content):
                    violations.append(rule['action'])
            elif rule['type'] == 'ai':
                # 使用AI判断
                check = self.llm.generate(
                    f"内容是否{rule['description']}？{content}"
                )
                if "是" in check:
                    violations.append(rule['action'])
        
        return violations