app.py

"""
ContentGuardian - Content Audit Agent
China · Simplified Chinese · Hanyu Xinxie Style · Text-based Version
Optimized for Hugging Face Spaces
"""
import gradio as gr
import datetime

def comprehensive_text_audit(text, keywords=""):
    """
    Comprehensive text audit - structured text output
    """
    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 []
    
    # 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                   ║
╚══════════════════════════════════════════════════════════════╝

📊 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')}

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🚫 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']))}
"""
    else:
        report += "\n   ✅ 未发现不当广告用语\n"
    
    report += f"""
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🔍 关键词标记 ({len(found_keywords)} 处)
"""
    
    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"
    
    report += f"""
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

📋 详细分析
├─ 合规建议：
"""
    
    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"
    
    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"}

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🎨 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。

╔══════════════════════════════════════════════════════════════╗
║  审核完成 - 感谢使用 ContentGuardian 内容审核智能体          ║
╚══════════════════════════════════════════════════════════════╝
"""
    
    return report

# 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.Textbox(
        label="Audit Report",
        lines=25,
        max_lines=30
    ),
    title="🛡️ ContentGuardian - Content Audit Agent",
    description="""
    **China · Simplified Chinese · Hanyu Xinxie Style · Text-based 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.
    
    **Key Features:**
    - 🚫 Inappropriate Advertising Language Detection (32+ common violation words)
    - 🔍 Precise Keyword Marking
    - 📊 Structured Audit Reports
    - 🎨 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"]
    ],
    theme=gr.themes.Soft(),
    flagging_mode="never"
)

if __name__ == "__main__":
    demo.launch()