Implement real AI-powered content audit with Hanyu Xinxie style cards

app.py

@@ -1,151 +1,381 @@
 """
-ContentGuardian - Content Audit Agent
-China · Simplified Chinese · Hanyu Xinxie Style · Text-based Version
-Optimized for Hugging Face Spaces
+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
 
-def comprehensive_text_audit(text, keywords=""):
-    """
-    Comprehensive text audit - structured text output
-    """
-    if not text.strip():
-        return "❌ Please enter text content"
+# 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"
     
-    # Parse keywords
-    keyword_list = [k.strip() for k in keywords.split(",") if k.strip()] if keywords else []
-    
-    # 1. Inappropriate advertising language detection
-    inappropriate_words = [
-        "绝对", "完全", "100%", "永远", "从不", "必须", "一定",
-        "第一", "最好", "最佳", "最优", "最强", "最大", "最小",
-        "极度", "超级", "顶级", "王牌", "冠军", "霸主",
-        "立即", "马上", "瞬间", "秒杀", "急速", "飞速",
-        "神奇", "奇迹", "万能", "无敌", "完美", "终极",
-        "absolutely", "completely", "perfect", "ultimate", "best", "first",
-        "immediately", "instantly", "magical", "miracle", "supreme"
-    ]
-    
-    found_inappropriate = []
-    for word in inappropriate_words:
-        if word.lower() in text.lower():
-            positions = []
-            start = 0
-            text_lower = text.lower()
-            word_lower = word.lower()
-            while True:
-                pos = text_lower.find(word_lower, start)
-                if pos == -1:
-                    break
-                positions.append(pos)
-                start = pos + 1
-            if positions:
-                found_inappropriate.append({
-                    "word": word,
-                    "count": len(positions),
-                    "positions": positions
-                })
-    
-    # 2. Keyword marking
-    found_keywords = []
-    for keyword in keyword_list:
-        if keyword.lower() in text.lower():
-            positions = []
-            start = 0
-            text_lower = text.lower()
-            keyword_lower = keyword.lower()
-            while True:
-                pos = text_lower.find(keyword_lower, start)
-                if pos == -1:
-                    break
-                positions.append(pos)
-                start = pos + 1
-            if positions:
-                found_keywords.append({
-                    "keyword": keyword,
-                    "count": len(positions),
-                    "positions": positions
-                })
-    
-    # 3. Generate structured report
-    total_issues = len(found_inappropriate)
-    severity = "低" if total_issues < 3 else "中" if total_issues < 6 else "高"
-    
-    # Build standardized text report
-    report = f"""
-╔══════════════════════════════════════════════════════════════╗
-║                    📋 内容审核报告                           ║
-║                  中国·简体中文·汉语新解风格                   ║
-╚══════════════════════════════════════════════════════════════╝
+    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：
 
-📊 审核概览
-├─ 总问题数：{total_issues} 处
-├─ 严重程度：{severity}
-├─ 文本长度：{len(text)} 字符
-└─ 审核时间：{datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')}
+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
 
-🚫 不当广告用语检测 ({len(found_inappropriate)} 处)
-"""
-    
-    if found_inappropriate:
-        for i, item in enumerate(found_inappropriate, 1):
-            report += f"""
-   {i}. 词汇："{item['word']}"
-      └─ 出现次数：{item['count']} 次
-      └─ 位置：{', '.join(map(str, item['positions']))}
+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"]
+    }}
+}}
 """
-    else:
-        report += "\n   ✅ 未发现不当广告用语\n"
+        
+        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)
     
-    report += f"""
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+    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"]
+            }
+        }
 
-🔍 关键词标记 ({len(found_keywords)} 处)
-"""
+# 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")
     
-    if found_keywords:
-        for i, item in enumerate(found_keywords, 1):
-            report += f"""
-   {i}. 关键词："{item['keyword']}"
-      └─ 出现次数：{item['count']} 次
-      └─ 位置：{', '.join(map(str, item['positions']))}
-"""
-    else:
-        report += "\n   ℹ️ 未指定关键词或关键词未出现\n"
+    # 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
+    }
     
-    report += f"""
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-📋 详细分析
-├─ 合规建议：
-"""
+    # 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
     
-    if found_inappropriate:
-        report += f"""│  • 建议修改或删除 {len(found_inappropriate)} 个不当用词
-│  • 使用相对性表述替代绝对化用词
-│  • 避免夸大宣传和误导性表述
-"""
-    else:
-        report += "│  • 内容表述规范，符合广告法要求\n"
+    # 生成主要摘要卡片
+    main_card = f"""
+    <div style="
+        display: flex; 
+        justify-content: center; 
+        padding: 20px;
+        background: linear-gradient(135deg, #E8E3DE 0%, #F2EDE9 100%);
+        border-radius: 15px;
+        font-family: 'Microsoft YaHei', 'Noto Sans SC', sans-serif;
+        margin-bottom: 20px;
+    ">
+        <div style="
+            width: 350px; 
+            background-color: {colors['background']}; 
+            border-radius: 20px; 
+            box-shadow: 0 20px 40px rgba(0,0,0,0.1); 
+            overflow: hidden; 
+            position: relative;
+        ">
+            <!-- 标题区域 -->
+            <div style="
+                background-color: {colors['secondary']}; 
+                color: {colors['background']}; 
+                padding: 20px; 
+                text-align: left;
+            ">
+                <h1 style="font-size: 20px; margin: 0; font-weight: 700;">🛡️ AI内容审核报告</h1>
+                <p style="font-size: 12px; margin: 5px 0 0 0; opacity: 0.9;">中国·简体中文·汉语新解风格</p>
+            </div>
+            
+            <!-- 内容区域 -->
+            <div style="padding: 25px 20px;">
+                <!-- 主要统计 -->
+                <div style="text-align: left; margin-bottom: 20px;">
+                    <div style="
+                        font-size: 36px; 
+                        color: {colors['text']}; 
+                        margin-bottom: 8px; 
+                        font-weight: bold;
+                        position: relative;
+                    ">
+                        {total_issues} 处问题
+                        <div style="
+                            position: absolute; 
+                            left: 0; 
+                            bottom: -4px; 
+                            width: 60px; 
+                            height: 3px; 
+                            background-color: {colors['accent']};
+                        "></div>
+                    </div>
+                    <div style="
+                        font-size: 16px; 
+                        color: {colors['light_text']}; 
+                        margin: 12px 0;
+                    ">风险等级: {risk_level}</div>
+                </div>
+                
+                <!-- 分隔线 -->
+                <div style="
+                    width: 100%; 
+                    height: 1px; 
+                    background-color: {colors['divider']}; 
+                    margin: 20px 0;
+                "></div>
+                
+                <!-- AI分析结果 -->
+                <div style="
+                    font-size: 14px; 
+                    line-height: 1.6; 
+                    text-align: left;
+                ">
+                    <div style="
+                        padding-left: 15px; 
+                        border-left: 3px solid {colors['accent']};
+                    ">"""
     
-    report += f"""├─ 风险等级：{severity}
-└─ 处理建议：{"需要重点关注和修改" if severity == "高" else "建议适当调整" if severity == "中" else "可以正常使用"}
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+    # 添加具体问题类型
+    if analysis_result.get("typos"):
+        main_card += f'<p style="margin: 8px 0; color: {colors["text"]};">• 错别字问题: {len(analysis_result["typos"])} 处</p>'
+    
+    if analysis_result.get("violations"):
+        main_card += f'<p style="margin: 8px 0; color: {colors["text"]};">• 违规内容: {len(analysis_result["violations"])} 处</p>'
+    
+    if analysis_result.get("fake_content"):
+        main_card += f'<p style="margin: 8px 0; color: {colors["text"]};">• 虚假内容: {len(analysis_result["fake_content"])} 处</p>'
+    
+    if analysis_result.get("inappropriate_ads"):
+        main_card += f'<p style="margin: 8px 0; color: {colors["text"]};">• 不当广告用语: {len(analysis_result["inappropriate_ads"])} 处</p>'
+    
+    if analysis_result.get("keywords_analysis"):
+        main_card += f'<p style="margin: 8px 0; color: {colors["text"]};">• 关键词分析: {len(analysis_result["keywords_analysis"])} 个</p>'
+    
+    if total_issues == 0:
+        main_card += f'<p style="margin: 8px 0; color: {colors["light_text"]};">✅ 未发现明显问题</p>'
+    
+    main_card += f"""
+                    </div>
+                </div>
+                
+                <!-- AI建议 -->
+                <div style="
+                    margin-top: 20px; 
+                    padding: 15px; 
+                    background-color: rgba(255,255,255,0.5); 
+                    border-radius: 10px;
+                    border-left: 4px solid {colors['accent']};
+                ">
+                    <h4 style="margin: 0 0 10px 0; color: {colors['text']}; font-size: 14px;">🤖 AI建议</h4>"""
+    
+    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'<p style="margin: 5px 0; color: {colors["light_text"]}; font-size: 13px;">• {rec}</p>'
+    
+    main_card += f"""
+                </div>
+                
+                <!-- 中式印章风格装饰 -->
+                <div style="
+                    text-align: center; 
+                    margin-top: 25px; 
+                    padding: 10px; 
+                    border: 2px solid {colors['accent']}; 
+                    border-radius: 50%; 
+                    width: 70px; 
+                    height: 70px; 
+                    margin-left: auto; 
+                    margin-right: auto;
+                    display: flex;
+                    align-items: center;
+                    justify-content: center;
+                ">
+                    <span style="
+                        font-size: 18px; 
+                        color: {colors['secondary']}; 
+                        font-weight: bold;
+                    ">AI审核</span>
+                </div>
+            </div>
+            
+            <!-- 背景装饰文字 -->
+            <div style="
+                position: absolute; 
+                font-size: 120px; 
+                color: rgba(182, 181, 167, 0.05); 
+                top: 50%; 
+                left: 50%; 
+                transform: translate(-50%, -50%); 
+                font-weight: bold;
+                pointer-events: none;
+                z-index: 0;
+            ">AI</div>
+        </div>
+    </div>
+    """
+    
+    return main_card
 
-🎨 设计理念
-本审核系统融合中华文化与现代设计，体现内敛典雅之美。
-采用汉语新解风格，注重文字的准确性和表达的规范性。
+def generate_error_card(error_msg: str) -> str:
+    """生成错误信息卡片"""
+    return f"""
+    <div style="
+        display: flex; 
+        justify-content: center; 
+        padding: 20px;
+        background: linear-gradient(135deg, #E8E3DE 0%, #F2EDE9 100%);
+        border-radius: 15px;
+        font-family: 'Microsoft YaHei', sans-serif;
+    ">
+        <div style="
+            width: 350px; 
+            background-color: #F2EDE9; 
+            border-radius: 20px; 
+            box-shadow: 0 20px 40px rgba(0,0,0,0.1); 
+            padding: 30px;
+            text-align: center;
+        ">
+            <h2 style="color: #B85450; margin-bottom: 15px;">⚠️ 分析出错</h2>
+            <p style="color: #5B5B5B; margin-bottom: 20px;">AI分析过程中遇到问题：</p>
+            <div style="
+                background-color: rgba(184, 84, 80, 0.1); 
+                padding: 15px; 
+                border-radius: 10px; 
+                border-left: 4px solid #B85450;
+                text-align: left;
+            ">
+                <code style="color: #B85450; font-size: 12px;">{error_msg}</code>
+            </div>
+            <p style="color: #8C8C8C; margin-top: 15px; font-size: 14px;">请稍后重试或检查网络连接</p>
+        </div>
+    </div>
+    """
 
-╔══════════════════════════════════════════════════════════════╗
-║  审核完成 - 感谢使用 ContentGuardian 内容审核智能体          ���
-╚══════════════════════════════════════════════════════════════╝
-"""
+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 []
     
-    return report
+    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(
@@ -162,32 +392,30 @@ demo = gr.Interface(
             lines=2
         )
     ],
-    outputs=gr.Textbox(
-        label="Audit Report",
-        lines=25,
-        max_lines=30
+    outputs=gr.HTML(
+        label="AI Audit Report"
     ),
-    title="🛡️ ContentGuardian - Content Audit Agent",
+    title="🛡️ ContentGuardian - AI Content Audit Agent",
     description="""
-    **China · Simplified Chinese · Hanyu Xinxie Style · Text-based Version**
+    **China · Simplified Chinese · Hanyu Xinxie Style · AI-Driven Version**
     
-    This system uses standardized structured text output, integrating Chinese culture with modern design concepts.
-    Detects inappropriate advertising language, marks specified keywords, and generates detailed audit reports.
+    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:**
-    - 🚫 Inappropriate Advertising Language Detection (32+ common violation words)
-    - 🔍 Precise Keyword Marking
-    - 📊 Structured Audit Reports
+    - 🤖 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"],
-        ["Buy now, instant effect, 100% effective, absolutely satisfying!", "buy,effect"],
-        ["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"]
+        ["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"
+    flagging_mode="never"
 )
 
 if __name__ == "__main__":

requirements.txt

@@ -1 +1,2 @@
 gradio>=4.0.0
+aiohttp>=3.8.0