添加rag置信度判断

backend/app/rag/pipeline.py

@@ -81,11 +81,17 @@ class RAGPipeline:
         return await self.retriever.ainvoke(query)
 
     async def _get_parents(self, child_docs: List[Document]) -> List[Document]:
-        parent_map = {}
+        # 收集 parent_id 和对应的分数
+        parent_map = {}  # parent_id -> (embedding_score, rerank_score)
+
         for doc in child_docs:
             pid = doc.metadata.get("parent_id")
             if pid and pid not in parent_map:
-                parent_map[pid] = doc.metadata.get("score", 0.0)
+                # embedding 分数
+                embedding_score = doc.metadata.get("score", 0.0)
+                # rerank 分数（如果有的话）
+                rerank_score = doc.metadata.get("rerank_score", 0.0)
+                parent_map[pid] = (embedding_score, rerank_score)
 
         if not parent_map:
             logger.warning("[Pipeline] 未找到 parent_id，返回子文档")
@@ -94,10 +100,19 @@ class RAGPipeline:
         try:
             from backend.rag_core import create_docstore
             docstore, _ = create_docstore()
-            # 同步获取（异步版本不存在）
             parent_docs = docstore.mget(list(parent_map.keys()))
-            parent_map2 = {d.metadata.get("id"): d for d in parent_docs if d}
-            result = [(parent_map2[pid], score) for pid, score in parent_map.items() if pid in parent_map2]
+
+            # 构建结果，保持分数信息
+            result = []
+            for doc in parent_docs:
+                if doc:
+                    pid = doc.metadata.get("id")
+                    scores = parent_map.get(pid, (0.0, 0.0))
+                    # 将分数添加到 metadata 中
+                    doc.metadata["embedding_score"] = scores[0]
+                    doc.metadata["rerank_score"] = scores[1]
+                    result.append((doc, scores[0] + scores[1] * 2))  # 综合分数，rerank 权重更高
+
             result.sort(key=lambda x: x[1], reverse=True)
             docs = [d for d, _ in result]
             logger.info(f"[Pipeline] 获取到 {len(docs)} 个父文档")