better：优化记忆系统，记忆系统现可以整合记忆

src/config/config.py

@@ -234,6 +234,10 @@ class BotConfig:
     forget_memory_interval: int = 600  # 记忆遗忘间隔（秒）
     memory_forget_time: int = 24  # 记忆遗忘时间（小时）
     memory_forget_percentage: float = 0.01  # 记忆遗忘比例
+    
+    consolidate_memory_interval: int = 1000  # 记忆整合间隔（秒）
+    consolidation_similarity_threshold: float = 0.7  # 相似度阈值
+    consolidate_memory_percentage: float = 0.01  # 检查节点比例
 
     memory_ban_words: list = field(
         default_factory=lambda: ["表情包", "图片", "回复", "聊天记录"]
@@ -594,6 +598,16 @@ class BotConfig:
                 config.build_memory_sample_length = memory_config.get(
                     "build_memory_sample_length", config.build_memory_sample_length
                 )
+            if config.INNER_VERSION in SpecifierSet(">=1.5.1"):
+                config.consolidate_memory_interval = memory_config.get(
+                    "consolidate_memory_interval", config.consolidate_memory_interval
+                )
+                config.consolidation_similarity_threshold = memory_config.get(
+                    "consolidation_similarity_threshold", config.consolidation_similarity_threshold
+                )
+                config.consolidate_memory_percentage = memory_config.get(
+                    "consolidate_memory_percentage", config.consolidate_memory_percentage
+                )
 
         def remote(parent: dict):
             remote_config = parent["remote"]

src/main.py

@@ -121,6 +121,7 @@ class MainSystem:
             tasks = [
                 self.build_memory_task(),
                 self.forget_memory_task(),
+                self.consolidate_memory_task(),
                 self.print_mood_task(),
                 self.remove_recalled_message_task(),
                 emoji_manager.start_periodic_check_register(),
@@ -145,6 +146,15 @@ class MainSystem:
             print("\033[1;32m[记忆遗忘]\033[0m 开始遗忘记忆...")
             await HippocampusManager.get_instance().forget_memory(percentage=global_config.memory_forget_percentage)
             print("\033[1;32m[记忆遗忘]\033[0m 记忆遗忘完成")
+            
+    @staticmethod
+    async def consolidate_memory_task():
+        """记忆整合任务"""
+        while True:
+            await asyncio.sleep(global_config.consolidate_memory_interval)
+            print("\033[1;32m[记忆整合]\033[0m 开始整合记忆...")
+            await HippocampusManager.get_instance().consolidate_memory()
+            print("\033[1;32m[记忆整合]\033[0m 记忆整合完成")
 
     async def print_mood_task(self):
         """打印情绪状态"""

src/plugins/heartFC_chat/heartflow_prompt_builder.py

@@ -221,9 +221,6 @@ class PromptBuilder:
             moderation_prompt=await global_prompt_manager.get_prompt_async("moderation_prompt"),
         )
 
-        prompt = await relationship_manager.convert_all_person_sign_to_person_name(prompt)
-        prompt = parse_text_timestamps(prompt, mode="lite")
-
         return prompt
 
     async def _build_prompt_normal(self, chat_stream, message_txt: str, sender_name: str = "某人") -> tuple[str, str]:

src/plugins/memory_system/Hippocampus.py

@@ -195,6 +195,7 @@ class Hippocampus:
         self.config = None
 
     def initialize(self, global_config):
+        # 使用导入的 MemoryConfig dataclass 和其 from_global_config 方法
         self.config = MemoryConfig.from_global_config(global_config)
         # 初始化子组件
         self.entorhinal_cortex = EntorhinalCortex(self)
@@ -237,7 +238,7 @@ class Hippocampus:
         # 不再需要 time_info 参数
         prompt = (
             f'这是一段文字：\n{text}\n\n我想让你基于这段文字来概括"{topic}"这个概念，帮我总结成一句自然的话，'
-            f"可以包含时间和人物，以及具体的观点。只输出这句话就好"
+            f"要求包含对这个概念的定义，内容，知识，可以包含时间和人物。只输出这句话就好"
         )
         return prompt
 
@@ -795,7 +796,7 @@ class EntorhinalCortex:
     def get_memory_sample(self):
         """从数据库获取记忆样本"""
         # 硬编码：每条消息最大记忆次数
-        max_memorized_time_per_msg = 3
+        max_memorized_time_per_msg = 2
 
         # 创建双峰分布的记忆调度器
         sample_scheduler = MemoryBuildScheduler(
@@ -1337,26 +1338,56 @@ class ParahippocampalGyrus:
         logger.info("[遗忘] 开始检查节点...")
         node_check_start = time.time()
         for node in nodes_to_check:
+            # 检查节点是否存在，以防在迭代中被移除（例如边移除导致）
+            if node not in self.memory_graph.G:
+                continue
+
             node_data = self.memory_graph.G.nodes[node]
+
+            # 首先获取记忆项
+            memory_items = node_data.get("memory_items", [])
+            if not isinstance(memory_items, list):
+                memory_items = [memory_items] if memory_items else []
+
+            # 新增：检查节点是否为空
+            if not memory_items:
+                try:
+                    self.memory_graph.G.remove_node(node)
+                    node_changes["removed"].append(f"{node}(空节点)") # 标记为空节点移除
+                    logger.debug(f"[遗忘] 移除了空的节点: {node}")
+                except nx.NetworkXError as e:
+                    logger.warning(f"[遗忘] 移除空节点 {node} 时发生错误（可能已被移除）: {e}")
+                continue # 处理下一个节点
+
+            # --- 如果节点不为空，则执行原来的不活跃检查和随机移除逻辑 ---
             last_modified = node_data.get("last_modified", current_time)
-
+            # 条件1：检查是否长时间未修改 (超过24小时)
             if current_time - last_modified > 3600 * 24:
-                memory_items = node_data.get("memory_items", [])
-                if not isinstance(memory_items, list):
-                    memory_items = [memory_items] if memory_items else []
-
+                # 条件2：再次确认节点包含记忆项（理论上已确认，但作为保险）
                 if memory_items:
                     current_count = len(memory_items)
-                    removed_item = random.choice(memory_items)
-                    memory_items.remove(removed_item)
+                    # 如果列表非空，才进行随机选择
+                    if current_count > 0:
+                        removed_item = random.choice(memory_items)
+                        try:
+                            memory_items.remove(removed_item)
 
-                    if memory_items:
-                        self.memory_graph.G.nodes[node]["memory_items"] = memory_items
-                        self.memory_graph.G.nodes[node]["last_modified"] = current_time
-                        node_changes["reduced"].append(f"{node} (数量: {current_count} -> {len(memory_items)})")
-                    else:
-                        self.memory_graph.G.remove_node(node)
-                        node_changes["removed"].append(node)
+                            # 条件3：检查移除后 memory_items 是否变空
+                            if memory_items: # 如果移除后列表不为空
+                                # self.memory_graph.G.nodes[node]["memory_items"] = memory_items # 直接修改列表即可
+                                self.memory_graph.G.nodes[node]["last_modified"] = current_time # 更新修改时间
+                                node_changes["reduced"].append(f"{node} (数量: {current_count} -> {len(memory_items)})")
+                            else: # 如果移除后列表为空
+                                # 尝试移除节点，处理可能的错误
+                                try:
+                                    self.memory_graph.G.remove_node(node)
+                                    node_changes["removed"].append(f"{node}(遗忘清空)") # 标记为遗忘清空
+                                    logger.debug(f"[遗忘] 节点 {node} 因移除最后一项而被清空。")
+                                except nx.NetworkXError as e:
+                                    logger.warning(f"[遗忘] 尝试移除节点 {node} 时发生错误（可能已被移除）：{e}")
+                        except ValueError:
+                            # 这个错误理论上不应发生，因为 removed_item 来自 memory_items
+                            logger.warning(f"[遗忘] 尝试从节点 '{node}' 移除不存在的项目 '{removed_item[:30]}...'")
         node_check_end = time.time()
         logger.info(f"[遗忘] 节点检查耗时: {node_check_end - node_check_start:.2f}秒")
 
@@ -1395,6 +1426,116 @@ class ParahippocampalGyrus:
         end_time = time.time()
         logger.info(f"[遗忘] 总耗时: {end_time - start_time:.2f}秒")
 
+    async def operation_consolidate_memory(self):
+        """整合记忆：合并节点内相似的记忆项"""
+        start_time = time.time()
+        percentage = self.config.consolidate_memory_percentage
+        similarity_threshold = self.config.consolidation_similarity_threshold
+        logger.info(f"[整合] 开始检查记忆节点... 检查比例: {percentage:.2%}, 合并阈值: {similarity_threshold}")
+
+        # 获取所有至少有2条记忆项的节点
+        eligible_nodes = []
+        for node, data in self.memory_graph.G.nodes(data=True):
+            memory_items = data.get("memory_items", [])
+            if isinstance(memory_items, list) and len(memory_items) >= 2:
+                eligible_nodes.append(node)
+
+        if not eligible_nodes:
+            logger.info("[整合] 没有找到包含多个记忆项的节点，无需整合。")
+            return
+
+        # 计算需要检查的节点数量
+        check_nodes_count = max(1, min(len(eligible_nodes), int(len(eligible_nodes) * percentage)))
+
+        # 随机抽取节点进行检查
+        try:
+            nodes_to_check = random.sample(eligible_nodes, check_nodes_count)
+        except ValueError as e:
+            logger.error(f"[整合] 抽样节点时出错: {e}")
+            return
+
+        logger.info(f"[整合] 将检查 {len(nodes_to_check)} / {len(eligible_nodes)} 个符合条件的节点。")
+
+        merged_count = 0
+        nodes_modified = set()
+        current_timestamp = datetime.datetime.now().timestamp()
+
+        for node in nodes_to_check:
+            node_data = self.memory_graph.G.nodes[node]
+            memory_items = node_data.get("memory_items", [])
+            if not isinstance(memory_items, list) or len(memory_items) < 2:
+                continue # 双重检查，理论上不会进入
+
+            items_copy = list(memory_items) # 创建副本以安全迭代和修改
+
+            # 遍历所有记忆项组合
+            for item1, item2 in combinations(items_copy, 2):
+                # 确保 item1 和 item2 仍然存在于原始列表中（可能已被之前的合并移除）
+                if item1 not in memory_items or item2 not in memory_items:
+                    continue
+
+                similarity = self._calculate_item_similarity(item1, item2)
+
+                if similarity >= similarity_threshold:
+                    logger.debug(f"[整合] 节点 '{node}' 中发现相似项 (相似度: {similarity:.2f}):")
+                    logger.trace(f"  - '{item1}'")
+                    logger.trace(f"  - '{item2}'")
+
+                    # 比较信息量
+                    info1 = calculate_information_content(item1)
+                    info2 = calculate_information_content(item2)
+
+                    if info1 >= info2:
+                        item_to_keep = item1
+                        item_to_remove = item2
+                    else:
+                        item_to_keep = item2
+                        item_to_remove = item1
+
+                    # 从原始列表中移除信息量较低的项
+                    try:
+                        memory_items.remove(item_to_remove)
+                        logger.info(f"[整合] 已合并节点 '{node}' 中的记忆，保留: '{item_to_keep[:60]}...', 移除: '{item_to_remove[:60]}...'" )
+                        merged_count += 1
+                        nodes_modified.add(node)
+                        node_data['last_modified'] = current_timestamp # 更新修改时间
+                        _merged_in_this_node = True
+                        break # 每个节点每次检查只合并一对
+                    except ValueError:
+                        # 如果项已经被移除（例如，在之前的迭代中作为 item_to_keep），则跳过
+                        logger.warning(f"[整合] 尝试移除节点 '{node}' 中不存在的项 '{item_to_remove[:30]}...'，可能已被合并。")
+                        continue
+            # # 如果节点内发生了合并，更新节点数据 (这种方式不安全，会丢失其他属性)        
+            # if merged_in_this_node:
+            #      self.memory_graph.G.nodes[node]["memory_items"] = memory_items
+
+
+        if merged_count > 0:
+            logger.info(f"[整合] 共合并了 {merged_count} 对相似记忆项，分布在 {len(nodes_modified)} 个节点中。")
+            sync_start = time.time()
+            logger.info("[整合] 开始将变更同步到数据库...")
+            # 使用 resync 更安全地处理删除和添加
+            await self.hippocampus.entorhinal_cortex.resync_memory_to_db()
+            sync_end = time.time()
+            logger.info(f"[整合] 数据库同步耗时: {sync_end - sync_start:.2f}秒")
+        else:
+            logger.info("[整合] 本次检查未发现需要合并的记忆项。")
+
+        end_time = time.time()
+        logger.info(f"[整合] 整合检查完成，总耗时: {end_time - start_time:.2f}秒")
+
+    @staticmethod
+    def _calculate_item_similarity(item1: str, item2: str) -> float:
+        """计算两条记忆项文本的余弦相似度"""
+        words1 = set(jieba.cut(item1))
+        words2 = set(jieba.cut(item2))
+        all_words = words1 | words2
+        if not all_words:
+            return 0.0
+        v1 = [1 if word in words1 else 0 for word in all_words]
+        v2 = [1 if word in words2 else 0 for word in all_words]
+        return cosine_similarity(v1, v2)
+
 
 class HippocampusManager:
     _instance = None
@@ -1433,12 +1574,12 @@ class HippocampusManager:
         edge_count = len(memory_graph.edges())
 
         logger.success(f"""--------------------------------
-                       记忆系统参数配置:
-                       构建间隔: {global_config.build_memory_interval}秒|样本数: {config.build_memory_sample_num},长度: {config.build_memory_sample_length}|压缩率: {config.memory_compress_rate}
-                       记忆构建分布: {config.memory_build_distribution}
-                       遗忘间隔: {global_config.forget_memory_interval}秒|遗忘比例: {global_config.memory_forget_percentage}|遗忘: {config.memory_forget_time}小时之后
-                       记忆图统计信息: 节点数量: {node_count}, 连接数量: {edge_count}
-                       --------------------------------""")  # noqa: E501
+                    记忆系统参数配置:
+                    构建间隔: {global_config.build_memory_interval}秒|样本数: {config.build_memory_sample_num},长度: {config.build_memory_sample_length}|压缩率: {config.memory_compress_rate}
+                    记忆构建分布: {config.memory_build_distribution}
+                    遗忘间隔: {global_config.forget_memory_interval}秒|遗忘比例: {global_config.memory_forget_percentage}|遗忘: {config.memory_forget_time}小时之后
+                    记忆图统计信息: 节点数量: {node_count}, 连接数量: {edge_count}
+                    --------------------------------""")  # noqa: E501
 
         return self._hippocampus
 
@@ -1453,6 +1594,14 @@ class HippocampusManager:
         if not self._initialized:
             raise RuntimeError("HippocampusManager 尚未初始化，请先调用 initialize 方法")
         return await self._hippocampus.parahippocampal_gyrus.operation_forget_topic(percentage)
+    
+    async def consolidate_memory(self):
+        """整合记忆的公共接口"""
+        if not self._initialized:
+            raise RuntimeError("HippocampusManager 尚未初始化，请先调用 initialize 方法")
+        # 注意：目前 operation_consolidate_memory 内部直接读取配置，percentage 参数暂时无效
+        # 如果需要外部控制比例，需要修改 operation_consolidate_memory
+        return await self._hippocampus.parahippocampal_gyrus.operation_consolidate_memory()
 
     async def get_memory_from_text(
         self,

src/plugins/memory_system/memory_config.py

@@ -18,19 +18,29 @@ class MemoryConfig:
     # 记忆过滤相关配置
     memory_ban_words: List[str]  # 记忆过滤词列表
 
+    # 新增：记忆整合相关配置
+    consolidation_similarity_threshold: float # 相似度阈值
+    consolidate_memory_percentage: float # 检查节点比例
+    consolidate_memory_interval: int # 记忆整合间隔
+
     llm_topic_judge: str  # 话题判断模型
     llm_summary_by_topic: str  # 话题总结模型
 
     @classmethod
     def from_global_config(cls, global_config):
         """从全局配置创建记忆系统配置"""
+        # 使用 getattr 提供默认值，防止全局配置缺少这些项
         return cls(
-            memory_build_distribution=global_config.memory_build_distribution,
-            build_memory_sample_num=global_config.build_memory_sample_num,
-            build_memory_sample_length=global_config.build_memory_sample_length,
-            memory_compress_rate=global_config.memory_compress_rate,
-            memory_forget_time=global_config.memory_forget_time,
-            memory_ban_words=global_config.memory_ban_words,
-            llm_topic_judge=global_config.llm_topic_judge,
-            llm_summary_by_topic=global_config.llm_summary_by_topic,
+            memory_build_distribution=getattr(global_config, "memory_build_distribution", (24, 12, 0.5, 168, 72, 0.5)), # 添加默认值
+            build_memory_sample_num=getattr(global_config, "build_memory_sample_num", 5),
+            build_memory_sample_length=getattr(global_config, "build_memory_sample_length", 30),
+            memory_compress_rate=getattr(global_config, "memory_compress_rate", 0.1),
+            memory_forget_time=getattr(global_config, "memory_forget_time", 24 * 7),
+            memory_ban_words=getattr(global_config, "memory_ban_words", []),
+            # 新增加载整合配置，并提供默认值
+            consolidation_similarity_threshold=getattr(global_config, "consolidation_similarity_threshold", 0.7),
+            consolidate_memory_percentage=getattr(global_config, "consolidate_memory_percentage", 0.01),
+            consolidate_memory_interval=getattr(global_config, "consolidate_memory_interval", 1000),
+            llm_topic_judge=getattr(global_config, "llm_topic_judge", "default_judge_model"), # 添加默认模型名
+            llm_summary_by_topic=getattr(global_config, "llm_summary_by_topic", "default_summary_model"), # 添加默认模型名
         )

src/plugins/person_info/relationship_manager.py

@@ -101,36 +101,6 @@ class RelationshipManager:
         # await person_info_manager.update_one_field(person_id, "user_avatar", user_avatar)
         await person_info_manager.qv_person_name(person_id, user_nickname, user_cardname, user_avatar)
 
-    @staticmethod
-    async def convert_all_person_sign_to_person_name(input_text: str):
-        """将所有人的<platform:user_id:nickname:cardname>格式转换为person_name"""
-        try:
-            # 使用正则表达式匹配<platform:user_id:nickname:cardname>格式
-            all_person = person_info_manager.person_name_list
-
-            pattern = r"<([^:]+):(\d+):([^:]+):([^>]+)>"
-            matches = re.findall(pattern, input_text)
-
-            # 遍历匹配结果，将<platform:user_id:nickname:cardname>替换为person_name
-            result_text = input_text
-            for platform, user_id, nickname, cardname in matches:
-                person_id = person_info_manager.get_person_id(platform, user_id)
-                # 默认使用昵称作为人名
-                person_name = nickname.strip() if nickname.strip() else cardname.strip()
-
-                if person_id in all_person:
-                    if all_person[person_id] is not None:
-                        person_name = all_person[person_id]
-
-                    # print(f"将<{platform}:{user_id}:{nickname}:{cardname}>替换为{person_name}")
-
-                result_text = result_text.replace(f"<{platform}:{user_id}:{nickname}:{cardname}>", person_name)
-
-            return result_text
-        except Exception:
-            logger.error(traceback.format_exc())
-            return input_text
-
     async def calculate_update_relationship_value(self, chat_stream: ChatStream, label: str, stance: str) -> tuple:
         """计算并变更关系值
         新的关系值变更计算方式：

template/bot_config_template.toml

@@ -1,5 +1,5 @@
 [inner]
-version = "1.5.0"
+version = "1.5.1"
 
 #----以下是给开发人员阅读的，如果你只是部署了麦麦，不需要阅读----
 #如果你想要修改配置文件，请在修改后将version的值进行变更
@@ -127,15 +127,19 @@ check_prompt = "符合公序良俗" # 表情包过滤要求，只有符合该要
 
 [memory]
 build_memory_interval = 2000 # 记忆构建间隔 单位秒   间隔越低，麦麦学习越多，但是冗余信息也会增多
-build_memory_distribution = [4.0,2.0,0.6,24.0,8.0,0.4] # 记忆构建分布，参数：分布1均值，标准差，权重，分布2均值，标准差，权重
-build_memory_sample_num = 10 # 采样数量，数值越高记忆采样次数越多
-build_memory_sample_length = 20 # 采样长度，数值越高一段记忆内容越丰富
+build_memory_distribution = [6.0,3.0,0.6,32.0,12.0,0.4] # 记忆构建分布，参数：分布1均值，标准差，权重，分布2均值，标准差，权重
+build_memory_sample_num = 8 # 采样数量，数值越高记忆采样次数越多
+build_memory_sample_length = 40 # 采样长度，数值越高一段记忆内容越丰富
 memory_compress_rate = 0.1 # 记忆压缩率 控制记忆精简程度 建议保持默认,调高可以获得更多信息，但是冗余信息也会增多
 
 forget_memory_interval = 1000 # 记忆遗忘间隔 单位秒   间隔越低，麦麦遗忘越频繁，记忆更精简，但更难学习
 memory_forget_time = 24 #多长时间后的记忆会被遗忘 单位小时 
 memory_forget_percentage = 0.01 # 记忆遗忘比例 控制记忆遗忘程度 越大遗忘越多 建议保持默认
 
+consolidate_memory_interval = 1000 # 记忆整合间隔 单位秒   间隔越低，麦麦整合越频繁，记忆更精简
+consolidation_similarity_threshold = 0.7 # 相似度阈值
+consolidation_check_percentage = 0.01 # 检查节点比例
+
 #不希望记忆的词，已经记忆的不会受到影响
 memory_ban_words = [ 
     # "403","张三"