refactor: 清理不需要的代码文件，添加.gitignore，优化项目结构

src/knowledge_base/knowledge_manager.py

@@ -125,16 +125,36 @@ class KnowledgeManager:
                     query_filter = query_filter.filter(KnowledgeEntry.is_verified == True)
                 
                 entries = query_filter.all()
+                # 若已验证为空，则回退到全部活跃条目
+                if not entries and verified_only:
+                    entries = session.query(KnowledgeEntry).filter(KnowledgeEntry.is_active == True).all()
                 
                 if not entries:
                     return []
                 
                 # 计算相似度
                 texts = [entry.question + " " + entry.answer for entry in entries]
-                query_vector = self.vectorizer.transform([query])
-                entry_vectors = self.vectorizer.transform(texts)
-                
-                similarities = cosine_similarity(query_vector, entry_vectors)[0]
+
+                # 确保向量器已训练
+                try:
+                    vocab_ok = hasattr(self.vectorizer, 'vocabulary_') and bool(self.vectorizer.vocabulary_)
+                    if not vocab_ok:
+                        self.vectorizer.fit(texts)
+                    query_vector = self.vectorizer.transform([query])
+                    entry_vectors = self.vectorizer.transform(texts)
+                    similarities = cosine_similarity(query_vector, entry_vectors)[0]
+                except Exception as vec_err:
+                    logger.warning(f"TF-IDF搜索失败，回退到子串匹配: {vec_err}")
+                    # 回退：子串匹配评分
+                    similarities = []
+                    q = query.strip()
+                    for t in texts:
+                        if not q:
+                            similarities.append(0.0)
+                        else:
+                            score = 1.0 if q in t else 0.0
+                            similarities.append(score)
+                    similarities = np.array(similarities, dtype=float)
                 
                 # 获取top_k个最相似的条目
                 top_indices = np.argsort(similarities)[-top_k:][::-1]