feat: RAG inline citations, source highlighting, and admin panel

- Increase RAG retrieval from 5 to 50 docs with relevance threshold (0.15) - LLM inline citations with [来源N] format and reference list - Clickable citation links scroll to source cards with highlight animation - Source previews with full chunk text and answer-matched highlighting - Message-scoped source IDs to fix cross-round citation targeting - Admin panel pages (knowledge, users, forum, course, settings) - Add rehype-raw dependency for HTML-in-markdown rendering Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-27 20:51:28 +08:00
parent 5d42a0573a
commit abcced35b8
23 changed files with 2249 additions and 112 deletions
@@ -0,0 +1,511 @@
+"""
+后台管理API
+"""
+from typing import List, Optional
+from datetime import datetime, timedelta
+
+from fastapi import APIRouter, Depends, HTTPException, status
+from pydantic import BaseModel
+from sqlalchemy.orm import Session
+from sqlalchemy import func, desc
+
+from ..core.database import get_db
+from ..core.security import get_current_user
+from ..models.user import User
+from ..models.chat import ChatSession, ChatMessage
+from ..models.document import Document, DocumentChunk
+from ..models.knowledge_base import KnowledgeBase
+from ..models.forum import ForumCategory, ForumPost, ForumReply
+
+router = APIRouter(prefix="/admin", tags=["后台管理"])
+
+
+async def require_admin(
+    current_user: str = Depends(get_current_user),
+    db: Session = Depends(get_db)
+) -> User:
+    """验证当前用户是否为管理员"""
+    user = db.query(User).filter(User.username == current_user).first()
+    if not user or not user.is_superuser:
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN,
+            detail="需要管理员权限"
+        )
+    return user
+
+
+# ===== 数据概览 =====
+
+class AdminDashboardStats(BaseModel):
+    total_users: int
+    active_users_7d: int
+    total_sessions: int
+    total_messages: int
+    total_documents: int
+    total_knowledge_bases: int
+    total_forum_posts: int
+    total_forum_replies: int
+    total_generated_images: int
+
+
+class AdminTrendItem(BaseModel):
+    date: str
+    count: int
+
+
+@router.get("/dashboard", response_model=AdminDashboardStats)
+async def get_dashboard(
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """获取管理后台仪表盘数据"""
+    total_users = db.query(func.count(User.id)).scalar() or 0
+
+    # 7天内活跃用户（有消息的用户）
+    week_ago = datetime.utcnow() - timedelta(days=7)
+    active_users_7d = db.query(func.count(func.distinct(ChatMessage.session_id))).join(
+        ChatSession
+    ).filter(
+        ChatMessage.created_at >= week_ago
+    ).scalar() or 0
+    # 更准确的计算：有消息的独立用户数
+    active_users_7d = db.query(func.count(func.distinct(ChatSession.user_id))).join(
+        ChatMessage
+    ).filter(
+        ChatMessage.created_at >= week_ago
+    ).scalar() or 0
+
+    total_sessions = db.query(func.count(ChatSession.id)).scalar() or 0
+    total_messages = db.query(func.count(ChatMessage.id)).scalar() or 0
+    total_documents = db.query(func.count(Document.id)).scalar() or 0
+    total_knowledge_bases = db.query(func.count(KnowledgeBase.id)).scalar() or 0
+    total_forum_posts = db.query(func.count(ForumPost.id)).scalar() or 0
+    total_forum_replies = db.query(func.count(ForumReply.id)).scalar() or 0
+
+    # 生成图像数（从目录统计）
+    from pathlib import Path
+    from ..core.config import get_settings
+    settings = get_settings()
+    generated_images_dir = Path(settings.generated_images_dir)
+    total_generated_images = 0
+    if generated_images_dir.exists():
+        image_extensions = {'.png', '.jpg', '.jpeg', '.webp', '.gif'}
+        total_generated_images = sum(
+            1 for f in generated_images_dir.iterdir()
+            if f.is_file() and f.suffix.lower() in image_extensions
+        )
+
+    return AdminDashboardStats(
+        total_users=total_users,
+        active_users_7d=active_users_7d,
+        total_sessions=total_sessions,
+        total_messages=total_messages,
+        total_documents=total_documents,
+        total_knowledge_bases=total_knowledge_bases,
+        total_forum_posts=total_forum_posts,
+        total_forum_replies=total_forum_replies,
+        total_generated_images=total_generated_images,
+    )
+
+
+@router.get("/trends/users")
+async def get_user_trends(
+    days: int = 30,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """获取用户注册趋势"""
+    start_date = datetime.utcnow() - timedelta(days=days)
+    results = db.query(
+        func.date(User.created_at).label("date"),
+        func.count(User.id).label("count")
+    ).filter(
+        User.created_at >= start_date
+    ).group_by(
+        func.date(User.created_at)
+    ).order_by("date").all()
+
+    return [{"date": str(r.date), "count": r.count} for r in results]
+
+
+@router.get("/trends/messages")
+async def get_message_trends(
+    days: int = 30,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """获取消息趋势"""
+    start_date = datetime.utcnow() - timedelta(days=days)
+    results = db.query(
+        func.date(ChatMessage.created_at).label("date"),
+        func.count(ChatMessage.id).label("count")
+    ).filter(
+        ChatMessage.created_at >= start_date
+    ).group_by(
+        func.date(ChatMessage.created_at)
+    ).order_by("date").all()
+
+    return [{"date": str(r.date), "count": r.count} for r in results]
+
+
+# ===== 用户管理 =====
+
+class AdminUserItem(BaseModel):
+    id: int
+    username: str
+    email: str
+    full_name: Optional[str]
+    is_active: bool
+    is_superuser: bool
+    created_at: Optional[str]
+    last_login: Optional[str]
+    session_count: int
+    message_count: int
+
+    class Config:
+        from_attributes = True
+
+
+@router.get("/users", response_model=List[AdminUserItem])
+async def list_users(
+    skip: int = 0,
+    limit: int = 50,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """获取用户列表"""
+    users = db.query(User).order_by(desc(User.id)).offset(skip).limit(limit).all()
+
+    result = []
+    for u in users:
+        session_count = db.query(func.count(ChatSession.id)).filter(
+            ChatSession.user_id == u.id
+        ).scalar() or 0
+        message_count = db.query(func.count(ChatMessage.id)).join(
+            ChatSession
+        ).filter(ChatSession.user_id == u.id).scalar() or 0
+
+        result.append(AdminUserItem(
+            id=u.id,
+            username=u.username,
+            email=u.email,
+            full_name=u.full_name,
+            is_active=u.is_active,
+            is_superuser=u.is_superuser,
+            created_at=u.created_at.isoformat() if u.created_at else None,
+            last_login=u.last_login.isoformat() if u.last_login else None,
+            session_count=session_count,
+            message_count=message_count,
+        ))
+    return result
+
+
+@router.put("/users/{user_id}/toggle-active")
+async def toggle_user_active(
+    user_id: int,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """启用/禁用用户"""
+    user = db.query(User).filter(User.id == user_id).first()
+    if not user:
+        raise HTTPException(status_code=404, detail="用户不存在")
+    if user.id == admin.id:
+        raise HTTPException(status_code=400, detail="不能禁用自己")
+
+    user.is_active = not user.is_active
+    db.commit()
+    return {"success": True, "is_active": user.is_active}
+
+
+@router.put("/users/{user_id}/toggle-admin")
+async def toggle_user_admin(
+    user_id: int,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """设为/取消管理员"""
+    user = db.query(User).filter(User.id == user_id).first()
+    if not user:
+        raise HTTPException(status_code=404, detail="用户不存在")
+    if user.id == admin.id:
+        raise HTTPException(status_code=400, detail="不能修改自己的管理员状态")
+
+    user.is_superuser = not user.is_superuser
+    db.commit()
+    return {"success": True, "is_superuser": user.is_superuser}
+
+
+@router.delete("/users/{user_id}")
+async def delete_user(
+    user_id: int,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """删除用户"""
+    user = db.query(User).filter(User.id == user_id).first()
+    if not user:
+        raise HTTPException(status_code=404, detail="用户不存在")
+    if user.id == admin.id:
+        raise HTTPException(status_code=400, detail="不能删除自己")
+
+    db.delete(user)
+    db.commit()
+    return {"success": True}
+
+
+# ===== 论坛管理 =====
+
+class AdminForumCategory(BaseModel):
+    id: int
+    slug: str
+    name: str
+    description: Optional[str]
+    post_count: int
+
+    class Config:
+        from_attributes = True
+
+
+class AdminForumCategoryCreate(BaseModel):
+    name: str
+    slug: str
+    description: Optional[str] = None
+
+
+@router.get("/forum/categories", response_model=List[AdminForumCategory])
+async def list_forum_categories(
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """获取论坛分类列表（管理）"""
+    categories = db.query(ForumCategory).order_by(ForumCategory.id).all()
+    result = []
+    for c in categories:
+        post_count = db.query(func.count(ForumPost.id)).filter(
+            ForumPost.category_id == c.id
+        ).scalar() or 0
+        result.append(AdminForumCategory(
+            id=c.id,
+            slug=c.slug,
+            name=c.name,
+            description=c.description,
+            post_count=post_count,
+        ))
+    return result
+
+
+@router.post("/forum/categories", response_model=AdminForumCategory)
+async def create_forum_category(
+    data: AdminForumCategoryCreate,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """创建论坛分类"""
+    existing = db.query(ForumCategory).filter(ForumCategory.slug == data.slug).first()
+    if existing:
+        raise HTTPException(status_code=400, detail="slug 已存在")
+    category = ForumCategory(
+        slug=data.slug,
+        name=data.name,
+        description=data.description,
+    )
+    db.add(category)
+    db.commit()
+    db.refresh(category)
+    return AdminForumCategory(
+        id=category.id,
+        slug=category.slug,
+        name=category.name,
+        description=category.description,
+        post_count=0,
+    )
+
+
+@router.put("/forum/categories/{category_id}")
+async def update_forum_category(
+    category_id: int,
+    data: AdminForumCategoryCreate,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """更新论坛分类"""
+    category = db.query(ForumCategory).filter(ForumCategory.id == category_id).first()
+    if not category:
+        raise HTTPException(status_code=404, detail="分类不存在")
+    category.name = data.name
+    category.slug = data.slug
+    category.description = data.description
+    db.commit()
+    return {"success": True}
+
+
+@router.delete("/forum/categories/{category_id}")
+async def delete_forum_category(
+    category_id: int,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """删除论坛分类"""
+    category = db.query(ForumCategory).filter(ForumCategory.id == category_id).first()
+    if not category:
+        raise HTTPException(status_code=404, detail="分类不存在")
+    # 删除该分类下的所有帖子和回复
+    posts = db.query(ForumPost).filter(ForumPost.category_id == category_id).all()
+    for post in posts:
+        db.query(ForumReply).filter(ForumReply.post_id == post.id).delete()
+    db.query(ForumPost).filter(ForumPost.category_id == category_id).delete()
+    db.delete(category)
+    db.commit()
+    return {"success": True}
+
+
+class AdminForumPost(BaseModel):
+    id: int
+    title: str
+    author_name: str
+    category_name: str
+    reply_count: int
+    created_at: str
+
+
+@router.get("/forum/posts", response_model=List[AdminForumPost])
+async def list_forum_posts(
+    skip: int = 0,
+    limit: int = 50,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """获取所有帖子列表（管理）"""
+    posts = db.query(ForumPost).order_by(desc(ForumPost.created_at)).offset(skip).limit(limit).all()
+    result = []
+    for p in posts:
+        author = db.query(User).filter(User.id == p.user_id).first()
+        category = db.query(ForumCategory).filter(ForumCategory.id == p.category_id).first()
+        reply_count = db.query(func.count(ForumReply.id)).filter(
+            ForumReply.post_id == p.id
+        ).scalar() or 0
+        result.append(AdminForumPost(
+            id=p.id,
+            title=p.title,
+            author_name=author.username if author else "未知",
+            category_name=category.name if category else "未知",
+            reply_count=reply_count,
+            created_at=p.created_at.isoformat() if p.created_at else "",
+        ))
+    return result
+
+
+@router.delete("/forum/posts/{post_id}")
+async def delete_forum_post(
+    post_id: int,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """删除帖子"""
+    post = db.query(ForumPost).filter(ForumPost.id == post_id).first()
+    if not post:
+        raise HTTPException(status_code=404, detail="帖子不存在")
+    db.query(ForumReply).filter(ForumReply.post_id == post_id).delete()
+    db.delete(post)
+    db.commit()
+    return {"success": True}
+
+
+# ===== 知识库管理 =====
+
+class AdminKnowledgeBase(BaseModel):
+    id: int
+    name: str
+    description: Optional[str]
+    owner_name: str
+    is_system: bool
+    document_count: int
+    chunk_count: int
+    created_at: str
+
+
+@router.get("/knowledge-bases", response_model=List[AdminKnowledgeBase])
+async def list_knowledge_bases(
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """获取所有知识库"""
+    kbs = db.query(KnowledgeBase).order_by(desc(KnowledgeBase.id)).all()
+    result = []
+    for kb in kbs:
+        owner = db.query(User).filter(User.id == kb.user_id).first()
+        doc_count = db.query(func.count(Document.id)).filter(
+            Document.knowledge_base_id == kb.id
+        ).scalar() or 0
+        chunk_count = db.query(func.count(DocumentChunk.id)).join(
+            Document
+        ).filter(Document.knowledge_base_id == kb.id).scalar() or 0
+        result.append(AdminKnowledgeBase(
+            id=kb.id,
+            name=kb.name,
+            description=kb.description,
+            owner_name=owner.username if owner else "系统",
+            is_system=kb.is_system,
+            document_count=doc_count,
+            chunk_count=chunk_count,
+            created_at=kb.created_at.isoformat() if kb.created_at else "",
+        ))
+    return result
+
+
+@router.delete("/knowledge-bases/{kb_id}")
+async def delete_knowledge_base(
+    kb_id: int,
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """删除知识库"""
+    kb = db.query(KnowledgeBase).filter(KnowledgeBase.id == kb_id).first()
+    if not kb:
+        raise HTTPException(status_code=404, detail="知识库不存在")
+    # 删除关联文档和chunks
+    docs = db.query(Document).filter(Document.knowledge_base_id == kb_id).all()
+    for doc in docs:
+        db.query(DocumentChunk).filter(DocumentChunk.document_id == doc.id).delete()
+    db.query(Document).filter(Document.knowledge_base_id == kb_id).delete()
+    db.delete(kb)
+    db.commit()
+    return {"success": True}
+
+
+# ===== 系统状态 =====
+
+@router.get("/system/status")
+async def get_system_status(
+    admin: User = Depends(require_admin),
+    db: Session = Depends(get_db)
+):
+    """获取系统状态"""
+    from ..core.config import get_settings
+    settings = get_settings()
+
+    # 数据库状态
+    db_ok = True
+    try:
+        db.execute(func.now())
+    except Exception:
+        db_ok = False
+
+    # 向量库状态
+    vector_store_ok = True
+    vector_count = 0
+    try:
+        from ..rag.vector_store import VectorStore
+        vs = VectorStore()
+        collection = vs.get_or_create_collection("knowledge_base")
+        vector_count = collection.count()
+    except Exception:
+        vector_store_ok = False
+
+    return {
+        "database": {"status": "ok" if db_ok else "error"},
+        "vector_store": {"status": "ok" if vector_store_ok else "error", "vector_count": vector_count},
+        "llm_model": settings.siliconflow_model,
+        "embedding_model": "text2vec-base-chinese",
+    }
@@ -242,7 +242,7 @@ async def stream_message(
                rag_chain = create_rag_chain(
                    knowledge_base_ids=request.knowledge_base_ids,
                    search_type="similarity",
-                    k=5,
+                    k=50,
                    model=request.model
                )
                
@@ -18,8 +18,8 @@ class RAGChain:
        self,
        knowledge_base_ids: Optional[List[int]] = None,
        search_type: str = "similarity",
-        k: int = 5,
-        score_threshold: float = 0.1,
+        k: int = 50,
+        score_threshold: float = 0.15,
        model: Optional[str] = None
    ):
        """初始化RAG链
@@ -58,8 +58,8 @@ class RAGChain:
        """创建检索链（LangChain 1.0 Runnable API）"""
        # 使用LangChain 1.0的runnable API构建RAG链
        def format_docs(docs):
-            """格式化文档"""
-            return "\n\n".join(doc.page_content for doc in docs)
+            """格式化文档（使用带编号的上下文格式）"""
+            return self._format_docs_for_context(docs)
        
        # 构建RAG链 - 只检索一次，返回完整结果
        def rag_with_sources(input_data):
@@ -115,7 +115,7 @@ class RAGChain:
        """流式调用（只流式输出答案）"""
        # 先获取文档（这是唯一一次检索）
        docs = await self.retriever.ainvoke(question)
-        context = "\n\n".join(doc.page_content for doc in docs)
+        context = self._format_docs_for_context(docs)
        
        # 构建prompt
        prompt_value = await self.prompt.ainvoke({"context": context, "question": question})
@@ -143,7 +143,7 @@ class RAGChain:
        # 发送检索结果（只在有文档时）
        if docs:
            doc_details = []
-            for i, doc in enumerate(docs[:5]):
+            for i, doc in enumerate(docs[:10]):
                metadata = doc.metadata if hasattr(doc, 'metadata') else {}
                title = metadata.get("title", metadata.get("filename", f"文档 {i+1}"))
                preview = doc.page_content[:100].replace('\n', ' ')
@@ -158,7 +158,7 @@ class RAGChain:
                "details": doc_details
            }

-        context = "\n\n".join(doc.page_content for doc in docs)
+        context = self._format_docs_for_context(docs)

        # 2. 构建prompt并流式生成
        prompt_value = await self.prompt.ainvoke({"context": context, "question": question})
@@ -190,20 +190,31 @@ class RAGChain:
    def _format_sources(self, documents: List) -> List[Dict]:
        """格式化来源信息"""
        sources = []
-        for doc in documents:
+        for i, doc in enumerate(documents, start=1):
            metadata = doc.metadata if hasattr(doc, 'metadata') else {}
            sources.append({
+                "id": i,
                "title": metadata.get("title", "未知标题"),
                "filename": metadata.get("filename", "未知文件"),
                "page": metadata.get("chunk_index", 0),
-                "preview": doc.page_content[:200] + "..." if len(doc.page_content) > 200 else doc.page_content
+                "score": metadata.get("score"),
+                "preview": doc.page_content
            })
        return sources

+    @staticmethod
+    def _format_docs_for_context(docs) -> str:
+        """将文档列表格式化为带编号的LLM上下文字符串"""
+        parts = []
+        for i, doc in enumerate(docs, start=1):
+            source_title = doc.metadata.get("title", doc.metadata.get("filename", "未知"))
+            parts.append(f"[来源{i}]（来源：{source_title}）\n{doc.page_content}")
+        return "\n\n".join(parts)
+
 def create_rag_chain(
    knowledge_base_ids: Optional[List[int]] = None,
    search_type: str = "similarity",
-    k: int = 5,
+    k: int = 50,
    model: Optional[str] = None
 ) -> RAGChain:
    """创建RAG链实例
@@ -7,15 +7,18 @@ from langchain_core.messages import SystemMessage, HumanMessage
 # RAG系统提示词
 RAG_SYSTEM_PROMPT = """你是一个专业的国土空间规划知识问答助手。请基于以下上下文信息回答用户的问题。

-上下文信息：
+上下文信息（每个来源已标注编号，如 [来源1]、[来源2]）：
 {context}

 要求：
-1. 回答要准确、专业、详细
-2. 如果上下文中没有相关信息，请诚实说明
-3. 回答要结构清晰，逻辑性强
-4. 适当引用相关概念和术语
-5. 回答长度控制在500-1000字之间
+1. 综合评估上下文信息的相关性，优先采用与问题最相关的内容，不强行引用不相关的来源
+2. 回答中必须使用 [来源N] 格式（N为上下文中的来源编号）标注所引用的具体来源，例如"根据[来源3]，国土空间规划..."
+3. 回答要准确、专业、详细，结构清晰，逻辑性强
+4. 如果上下文中没有相关信息，请诚实说明
+5. 在回答末尾，列出所有实际引用的参考来源，格式为：
+   **参考来源：**
+   - [来源N] 文档标题
+6. 回答长度控制在500-1000字之间

 请基于上述上下文信息回答用户的问题。"""

@@ -14,7 +14,7 @@ class KnowledgeBaseRetriever(BaseRetriever):
    knowledge_base_ids: Optional[List[int]] = None
    search_type: str = "similarity"
    search_kwargs: dict = {"k": 5}
-    score_threshold: float = 0.1
+    score_threshold: float = 0.15
    
    def _get_relevant_documents(
        self,