From c5b9bc4927e1d35f5099c361444da79696706a9f Mon Sep 17 00:00:00 2001
From: SengokuCola <1026294844@qq.com>
Date: Sat, 25 Oct 2025 23:58:58 +0800
Subject: [PATCH] =?UTF-8?q?better=EF=BC=9A=E4=BC=98=E5=8C=96=E8=AE=B0?=
 =?UTF-8?q?=E5=BF=86=E6=9E=84=E5=BB=BA=E5=92=8C=E5=90=88=E5=B9=B6?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 src/memory_system/Memory_chest.py           | 114 ++++++++++----------
 src/memory_system/memory_management_task.py |   2 +-
 src/memory_system/memory_utils.py           |  53 ++++++++-
 src/plugins/built_in/memory/build_memory.py |  20 ++++
 4 files changed, 131 insertions(+), 58 deletions(-)

diff --git a/src/memory_system/Memory_chest.py b/src/memory_system/Memory_chest.py
index e28267c0..bd1d0a1c 100644
--- a/src/memory_system/Memory_chest.py
+++ b/src/memory_system/Memory_chest.py
@@ -18,7 +18,7 @@ from .memory_utils import (
     find_best_matching_memory,
     check_title_exists_fuzzy,
     get_all_titles,
-    get_memory_titles_by_chat_id_weighted,
+    find_most_similar_memory_by_chat_id,
 
 )
 
@@ -430,75 +430,40 @@ class MemoryChest:
         except Exception as e:
             logger.error(f"保存记忆仓库内容时出错: {e}")
     
-    async def choose_merge_target(self, memory_title: str, chat_id: str = None) -> list[str]:
+    async def choose_merge_target(self, memory_title: str, chat_id: str = None) -> tuple[list[str], list[str]]:
         """
-        选择与给定记忆标题相关的记忆目标
+        选择与给定记忆标题相关的记忆目标（基于文本相似度）
 
         Args:
             memory_title: 要匹配的记忆标题
-            chat_id: 聊天ID，用于加权抽样
+            chat_id: 聊天ID，用于筛选同chat_id的记忆
 
         Returns:
-            list[str]: 选中的记忆内容列表
+            tuple[list[str], list[str]]: (选中的记忆标题列表, 选中的记忆内容列表)
         """
         try:
-        # 如果提供了chat_id，使用加权抽样
-            all_titles = get_memory_titles_by_chat_id_weighted(chat_id)
-            # 剔除掉输入的 memory_title 本身
-            all_titles = [title for title in all_titles if title and title.strip() != (memory_title or "").strip()]
+            if not chat_id:
+                logger.warning("未提供chat_id，无法进行记忆匹配")
+                return [], []
             
-            content = ""
-            display_index = 1
-            for title in all_titles:
-                content += f"{display_index}. {title}\n"
-                display_index += 1
+            # 使用相似度匹配查找最相似的记忆
+            similar_memory = find_most_similar_memory_by_chat_id(
+                target_title=memory_title,
+                target_chat_id=chat_id,
+                similarity_threshold=0.5  # 相似度阈值
+            )
             
-            prompt = f"""
-所有记忆列表
-{content}
-
-请根据以上记忆列表，选择一个与"{memory_title}"相关的记忆，用json输出：
-如果没有相关记忆，输出:
-{{
-    "selected_title": ""
-}}
-可以选择多个相关的记忆，但最多不超过5个
-例如：
-{{
-    "selected_title": "选择的相关记忆标题"
-}},
-{{
-    "selected_title": "选择的相关记忆标题"
-}},
-{{
-    "selected_title": "选择的相关记忆标题"
-}}
-...
-注意：请返回原始标题本身作为 selected_title，不要包含前面的序号或多余字符。
-请输出JSON格式，不要输出其他内容：
-"""
-
-            # logger.info(f"选择合并目标 prompt: {prompt}")
-
-            if global_config.debug.show_prompt:
-                logger.info(f"选择合并目标 prompt: {prompt}")
+            if similar_memory:
+                selected_title, selected_content, similarity = similar_memory
+                logger.info(f"为 '{memory_title}' 找到相似记忆: '{selected_title}' (相似度: {similarity:.3f})")
+                return [selected_title], [selected_content]
             else:
-                logger.debug(f"选择合并目标 prompt: {prompt}")
-                
-            merge_target_response, (reasoning_content, model_name, tool_calls) = await self.LLMRequest_build.generate_response_async(prompt)
-            
-            # 解析JSON响应
-            selected_titles = self._parse_merge_target_json(merge_target_response)
-            
-            # 根据标题查找对应的内容
-            selected_contents = self._get_memories_by_titles(selected_titles)
-            
-            logger.info(f"选择合并目标结果: {len(selected_contents)} 条记忆:{selected_titles}")
-            return selected_titles,selected_contents
+                logger.info(f"为 '{memory_title}' 未找到相似度 >= 0.7 的记忆")
+                return [], []
             
         except Exception as e:
             logger.error(f"选择合并目标时出错: {e}")
-            return []
+            return [], []
     
     def _get_memories_by_titles(self, titles: list[str]) -> list[str]:
         """
@@ -659,6 +624,11 @@ class MemoryChest:
         合并记忆
         """
         try:
+            # 在记忆整合前先清理空chat_id的记忆
+            cleaned_count = self.cleanup_empty_chat_id_memories()
+            if cleaned_count > 0:
+                logger.info(f"记忆整合前清理了 {cleaned_count} 条空chat_id记忆")
+            
             content = ""
             for memory in memory_list:
                 content += f"{memory}\n"
@@ -793,5 +763,37 @@ MutePlugin 是禁言插件的名称
             logger.error(f"生成合并记忆标题时出错: {e}")
             return f"合并记忆_{int(time.time())}"
     
+    def cleanup_empty_chat_id_memories(self) -> int:
+        """
+        清理chat_id为空的记忆记录
+        
+        Returns:
+            int: 被清理的记忆数量
+        """
+        try:
+            # 查找所有chat_id为空的记忆
+            empty_chat_id_memories = MemoryChestModel.select().where(
+                (MemoryChestModel.chat_id.is_null()) | 
+                (MemoryChestModel.chat_id == "") |
+                (MemoryChestModel.chat_id == "None")
+            )
+            
+            count = 0
+            for memory in empty_chat_id_memories:
+                logger.info(f"清理空chat_id记忆: 标题='{memory.title}', ID={memory.id}")
+                memory.delete_instance()
+                count += 1
+            
+            if count > 0:
+                logger.info(f"已清理 {count} 条chat_id为空的记忆记录")
+            else:
+                logger.debug("未发现需要清理的空chat_id记忆记录")
+                
+            return count
+            
+        except Exception as e:
+            logger.error(f"清理空chat_id记忆时出错: {e}")
+            return 0
+    
     
 global_memory_chest = MemoryChest()
\ No newline at end of file
diff --git a/src/memory_system/memory_management_task.py b/src/memory_system/memory_management_task.py
index 4d89658d..d750c1e5 100644
--- a/src/memory_system/memory_management_task.py
+++ b/src/memory_system/memory_management_task.py
@@ -109,7 +109,7 @@ class MemoryManagementTask(AsyncTask):
                 logger.info("无合适合并内容，跳过本次合并")
                 return
             
-            logger.info(f"为 [{selected_title}] 找到 {len(related_contents)} 条相关记忆:{related_titles}")
+            logger.info(f"{selected_chat_id} 为 [{selected_title}] 找到 {len(related_contents)} 条相关记忆:{related_titles}")
             
             # 执行merge_memory合并记忆
             merged_title, merged_content = await global_memory_chest.merge_memory(related_contents,selected_chat_id)
diff --git a/src/memory_system/memory_utils.py b/src/memory_system/memory_utils.py
index bdf8c2d3..bc5347bb 100644
--- a/src/memory_system/memory_utils.py
+++ b/src/memory_system/memory_utils.py
@@ -303,4 +303,55 @@ def get_memory_titles_by_chat_id_weighted(target_chat_id: str, same_chat_weight:
         
     except Exception as e:
         logger.error(f"按chat_id加权抽样记忆标题时出错: {e}")
-        return []
\ No newline at end of file
+        return []
+
+
+def find_most_similar_memory_by_chat_id(target_title: str, target_chat_id: str, similarity_threshold: float = 0.5) -> Optional[Tuple[str, str, float]]:
+    """
+    在指定chat_id的记忆中查找最相似的记忆
+    
+    Args:
+        target_title: 目标标题
+        target_chat_id: 目标聊天ID
+        similarity_threshold: 相似度阈值，默认0.7
+        
+    Returns:
+        Optional[Tuple[str, str, float]]: 最相似的记忆(title, content, similarity)或None
+    """
+    try:
+        # 获取指定chat_id的所有记忆
+        same_chat_memories = []
+        for memory in MemoryChestModel.select():
+            if memory.title and not memory.locked and memory.chat_id == target_chat_id:
+                same_chat_memories.append((memory.title, memory.content))
+        
+        if not same_chat_memories:
+            logger.warning(f"未找到chat_id为 '{target_chat_id}' 的记忆")
+            return None
+        
+        # 计算相似度并找到最佳匹配
+        best_match = None
+        best_similarity = 0.0
+        
+        for title, content in same_chat_memories:
+            # 跳过目标标题本身
+            if title.strip() == target_title.strip():
+                continue
+                
+            similarity = calculate_similarity(target_title, title)
+            
+            if similarity > best_similarity:
+                best_similarity = similarity
+                best_match = (title, content, similarity)
+        
+        # 检查是否超过阈值
+        if best_match and best_similarity >= similarity_threshold:
+            logger.info(f"找到最相似记忆: '{best_match[0]}' (相似度: {best_similarity:.3f})")
+            return best_match
+        else:
+            logger.info(f"未找到相似度 >= {similarity_threshold} 的记忆，最高相似度: {best_similarity:.3f}")
+            return None
+            
+    except Exception as e:
+        logger.error(f"查找最相似记忆时出错: {e}")
+        return None
\ No newline at end of file
diff --git a/src/plugins/built_in/memory/build_memory.py b/src/plugins/built_in/memory/build_memory.py
index 02e71447..11dcbebc 100644
--- a/src/plugins/built_in/memory/build_memory.py
+++ b/src/plugins/built_in/memory/build_memory.py
@@ -146,6 +146,20 @@ class GetMemoryTool(BaseTool):
         """从聊天记录中获取问题的答案"""
         try:
             # 确定时间范围
+            print(f"time_point: {time_point}, time_range: {time_range}")
+            
+            # 检查time_range的两个时间值是否相同，如果相同则按照time_point处理
+            if time_range and not time_point:
+                try:
+                    start_timestamp, end_timestamp = parse_time_range(time_range)
+                    if start_timestamp == end_timestamp:
+                        # 两个时间值相同，按照time_point处理
+                        time_point = time_range.split(" - ")[0].strip()
+                        time_range = None
+                        print(f"time_range两个值相同，按照time_point处理: {time_point}")
+                except Exception as e:
+                    logger.warning(f"解析time_range失败: {e}")
+            
             if time_point:
                 # 时间点：搜索前后25条记录
                 target_timestamp = parse_datetime_to_timestamp(time_point)
@@ -204,12 +218,18 @@ class GetMemoryTool(BaseTool):
 
 答案："""
                 
+                print(f"analysis_prompt: {analysis_prompt}")
+                
+                
                 response, (reasoning, model_name, tool_calls) = await llm_request.generate_response_async(
                     prompt=analysis_prompt,
                     temperature=0.3,
                     max_tokens=256
                 )
                 
+                
+                print(f"response: {response}")
+                
                 if "无有效信息" in response:
                     return ""