""" ContentGuardian - AI-Powered Content Audit Agent China · Simplified Chinese · Hanyu Xinxie Style · AI-Driven Version Real AI content analysis with SVG visualization """ import gradio as gr import asyncio import aiohttp import json import os import datetime from typing import Dict, List # AI Client for content analysis class AIContentAnalyzer: def __init__(self): self.api_key = os.getenv("SILICONFLOW_API_KEY", "sk-your-api-key-here") self.base_url = "https://api.siliconflow.cn/v1/chat/completions" self.model = "Qwen/Qwen2.5-7B-Instruct" async def analyze_content(self, text: str, keywords: List[str] = None) -> Dict: """AI-powered content analysis""" keywords_str = ", ".join(keywords) if keywords else "无" prompt = f""" 你是专业的内容审核专家，请对以下文本进行全面审核分析： 文本内容：{text} 关键词：{keywords_str} 请从以下维度进行分析： 1. 错别字检测：识别文本中的拼写错误和用词不当 2. 违规内容：检测不当言论、违法违规表述 3. 虚假内容：识别可能的虚假信息和夸大宣传 4. 不当广告用语：检测违反广告法的用词 5. 关键词分析：分析指定关键词的使用情况 请以JSON格式返回分析结果： {{ "typos": [ {{"original": "错误词", "correct": "正确词", "reason": "错误原因"}} ], "violations": [ {{"content": "违规内容", "type": "违规类型", "suggestion": "修改建议"}} ], "fake_content": [ {{"content": "虚假内容", "type": "虚假类型", "suggestion": "修改建议"}} ], "inappropriate_ads": [ {{"word": "不当用词", "reason": "违规原因", "suggestion": "替代建议"}} ], "keywords_analysis": [ {{"keyword": "关键词", "frequency": 次数, "context": "使用语境", "assessment": "使用评价"}} ], "overall_assessment": {{ "risk_level": "低/中/高", "total_issues": 问题总数, "main_concerns": ["主要问题1", "主要问题2"], "recommendations": ["建议1", "建议2"] }} }} """ try: headers = { "Authorization": f"Bearer {self.api_key}", "Content-Type": "application/json" } data = { "model": self.model, "messages": [ {"role": "system", "content": "你是专业的内容审核专家，擅长识别各类内容问题并提供专业建议。"}, {"role": "user", "content": prompt} ], "temperature": 0.1, "max_tokens": 2000 } async with aiohttp.ClientSession() as session: async with session.post(self.base_url, headers=headers, json=data) as response: if response.status == 200: result = await response.json() ai_response = result["choices"][0]["message"]["content"] # 尝试解析JSON try: # 清理AI返回的内容 if "```json" in ai_response: ai_response = ai_response.split("```json")[1].split("```")[0] elif "```" in ai_response: ai_response = ai_response.split("```")[1].split("```")[0] analysis_result = json.loads(ai_response.strip()) return analysis_result except json.JSONDecodeError: # 如果JSON解析失败，返回基础分析 return self._create_fallback_analysis(text, keywords) else: return self._create_fallback_analysis(text, keywords) except Exception as e: print(f"AI analysis error: {e}") return self._create_fallback_analysis(text, keywords) def _create_fallback_analysis(self, text: str, keywords: List[str] = None) -> Dict: """备用分析方法""" # 简单的关键词检测作为备用 inappropriate_words = ["绝对", "完全", "100%", "第一", "最好", "立即", "马上"] found_inappropriate = [word for word in inappropriate_words if word in text] return { "typos": [], "violations": [], "fake_content": [], "inappropriate_ads": [{"word": word, "reason": "可能违反广告法", "suggestion": "使用相对性表述"} for word in found_inappropriate], "keywords_analysis": [{"keyword": kw, "frequency": text.count(kw), "context": "文本中", "assessment": "正常使用"} for kw in (keywords or []) if kw in text], "overall_assessment": { "risk_level": "中" if found_inappropriate else "低", "total_issues": len(found_inappropriate), "main_concerns": ["广告用语不当"] if found_inappropriate else [], "recommendations": ["修改绝对化表述"] if found_inappropriate else ["内容表述规范"] } } # 全局AI分析器 ai_analyzer = AIContentAnalyzer() def generate_hanyu_xinxie_cards(analysis_result: Dict, original_text: str) -> str: """生成汉语新解风格的SVG卡片""" overall = analysis_result.get("overall_assessment", {}) total_issues = overall.get("total_issues", 0) risk_level = overall.get("risk_level", "低") # 莫兰迪色系配色 colors = { "primary": "#B6B5A7", # 莫兰迪灰褐色 "secondary": "#9A8F8F", # 莫兰迪灰棕色 "accent": "#C5B4A0", # 莫兰迪淡棕色 "background": "#F2EDE9", # 莫兰迪浅米色 "text": "#5B5B5B", # 莫兰迪深灰色 "light_text": "#8C8C8C", # 莫兰迪中灰色 "divider": "#D1CBC3" # 莫兰迪浅灰色 } # 根据风险等级调整颜色 if risk_level == "高": colors["secondary"] = "#B85450" # 深红色 colors["accent"] = "#D4776B" # 浅红色 elif risk_level == "中": colors["secondary"] = "#C4965A" # 橙色 colors["accent"] = "#D4A574" # 浅橙色 # 生成主要摘要卡片 main_card = f"""

🛡️ AI内容审核报告

中国·简体中文·汉语新解风格

{total_issues} 处问题
风险等级: {risk_level}
""" # 添加具体问题类型 if analysis_result.get("typos"): main_card += f'

• 错别字问题: {len(analysis_result["typos"])} 处

' if analysis_result.get("violations"): main_card += f'

• 违规内容: {len(analysis_result["violations"])} 处

' if analysis_result.get("fake_content"): main_card += f'

• 虚假内容: {len(analysis_result["fake_content"])} 处

' if analysis_result.get("inappropriate_ads"): main_card += f'

• 不当广告用语: {len(analysis_result["inappropriate_ads"])} 处

' if analysis_result.get("keywords_analysis"): main_card += f'

• 关键词分析: {len(analysis_result["keywords_analysis"])} 个

' if total_issues == 0: main_card += f'

✅ 未发现明显问题

' main_card += f"""

🤖 AI建议

""" recommendations = overall.get("recommendations", ["内容表述规范"]) for rec in recommendations[:2]: # 最多显示2个建议 main_card += f'

• {rec}

' main_card += f"""
AI审核
AI
""" return main_card def generate_error_card(error_msg: str) -> str: """生成错误信息卡片""" return f"""

⚠️ 分析出错

AI分析过程中遇到问题：

{error_msg}

请稍后重试或检查网络连接

""" def comprehensive_text_audit(text, keywords=""): """ AI-powered comprehensive text audit with SVG visualization """ if not text.strip(): return "❌ Please enter text content" # Parse keywords keyword_list = [k.strip() for k in keywords.split(",") if k.strip()] if keywords else [] try: # 使用异步事件循环进行AI分析 loop = asyncio.new_event_loop() asyncio.set_event_loop(loop) try: # 执行AI分析 analysis_result = loop.run_until_complete( ai_analyzer.analyze_content(text, keyword_list) ) finally: loop.close() # 生成汉语新解风格的SVG卡片 svg_cards = generate_hanyu_xinxie_cards(analysis_result, text) return svg_cards except Exception as e: print(f"Analysis error: {e}") # 返回错误信息的SVG卡片 return generate_error_card(str(e)) # Create Gradio interface optimized for HF Spaces demo = gr.Interface( fn=comprehensive_text_audit, inputs=[ gr.Textbox( label="Text to Audit", placeholder="Please enter the text content to be audited...", lines=8 ), gr.Textbox( label="Keywords (Optional)", placeholder="Please enter keywords to mark, separated by commas", lines=2 ) ], outputs=gr.HTML( label="AI Audit Report" ), title="🛡️ ContentGuardian - AI Content Audit Agent", description=""" **China · Simplified Chinese · Hanyu Xinxie Style · AI-Driven Version** This system uses real AI analysis to detect content issues and generates beautiful Hanyu Xinxie style cards. Powered by advanced language models for intelligent content understanding. **Key Features:** - 🤖 AI-Powered Content Analysis - 🔍 Intelligent Error Detection - 📊 Beautiful SVG Card Visualization - 🎨 Hanyu Xinxie Aesthetic Style """, examples=[ ["This product is absolutely effective, completely side-effect free, the first brand! Buy immediately, instant results!", "product,effect"], ["Our product quality is very good, trustworthy, welcome to purchase.", "product,quality"], ["这款产品效果绝对好，完全无副作用，第一品牌！立即购买，马上见效！", "产品,效果"], ["Buy now, instant effect, 100% effective, absolutely satisfying!", "buy,effect"] ], theme=gr.themes.Soft(), allow_flagging="never" ) if __name__ == "__main__": demo.launch()