From 4ae8a69fb7a0cd1c9e0efbc8057fb48506356b58 Mon Sep 17 00:00:00 2001 From: Cindy-Master <98833590+Cindy-Master@users.noreply.github.com> Date: Wed, 12 Mar 2025 09:51:49 +0800 Subject: [PATCH] =?UTF-8?q?=E5=86=8D=E6=AC=A1=E5=B0=9D=E8=AF=95=E5=AF=B9?= =?UTF-8?q?=E5=90=8C=E4=B9=89=E8=AF=8D=E6=B7=B7=E6=B7=86=E7=9A=84=E4=BC=98?= =?UTF-8?q?=E5=8C=96=20=20=E4=BC=9A=E4=BC=98=E5=85=88=E7=BE=A4=E8=81=8A?= =?UTF-8?q?=E5=BD=93=E5=89=8D=E8=AE=B0=E5=BF=86?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- src/plugins/memory_system/memory.py | 193 +++++++++++++++++++--------- 1 file changed, 135 insertions(+), 58 deletions(-) diff --git a/src/plugins/memory_system/memory.py b/src/plugins/memory_system/memory.py index 700c0957..6749ca9c 100644 --- a/src/plugins/memory_system/memory.py +++ b/src/plugins/memory_system/memory.py @@ -281,23 +281,32 @@ class Hippocampus: # 等待所有任务完成 compressed_memory = set() similar_topics_dict = {} # 存储每个话题的相似主题列表 + for topic, task in tasks: response = await task if response: - # 使用字典结构存储记忆内容与群组ID - memory_content = { - 'content': response[0], - 'group_id': group_id - } - compressed_memory.add((topic, memory_content)) + # 为每个主题创建特定群聊的主题节点名称 + group_specific_topic = topic + if group_id: + group_specific_topic = f"{topic}_g{group_id}" - # 为每个话题查找相似的已存在主题 + # 记录主题内容 + memory_content = response[0] + compressed_memory.add((group_specific_topic, memory_content)) + + # 为每个话题查找相似的已存在主题（不考虑群聊前缀） existing_topics = list(self.memory_graph.G.nodes()) similar_topics = [] for existing_topic in existing_topics: + # 提取基础主题名（去除群聊前缀） + base_existing_topic = existing_topic + if "_g" in existing_topic: + base_existing_topic = existing_topic.split("_g")[0] + + # 计算基础主题的相似度 topic_words = set(jieba.cut(topic)) - existing_words = set(jieba.cut(existing_topic)) + existing_words = set(jieba.cut(base_existing_topic)) all_words = topic_words | existing_words v1 = [1 if word in topic_words else 0 for word in all_words] @@ -305,12 +314,13 @@ class Hippocampus: similarity = cosine_similarity(v1, v2) - if similarity >= 0.6: + # 如果相似度高且不是同一主题在同一群聊 + if similarity >= 0.6 and existing_topic != group_specific_topic: similar_topics.append((existing_topic, similarity)) similar_topics.sort(key=lambda x: x[1], reverse=True) similar_topics = similar_topics[:5] - similar_topics_dict[topic] = similar_topics + similar_topics_dict[group_specific_topic] = similar_topics return compressed_memory, similar_topics_dict @@ -351,15 +361,25 @@ class Hippocampus: for topic, memory in compressed_memory: logger.info(f"添加节点: {topic}") - self.memory_graph.add_dot(topic, memory, group_id) + self.memory_graph.add_dot(topic, memory) # 不再需要传递group_id，因为已经包含在topic名称中 all_topics.append(topic) - # 连接相似的已存在主题 + # 连接相似的已存在主题，但使用较弱的连接强度 if topic in similar_topics_dict: similar_topics = similar_topics_dict[topic] for similar_topic, similarity in similar_topics: if topic != similar_topic: + # 如果是跨群聊的相似主题，使用较弱的连接强度 + is_cross_group = False + if ("_g" in topic and "_g" in similar_topic and + topic.split("_g")[1] != similar_topic.split("_g")[1]): + is_cross_group = True + + # 跨群聊的相似主题使用较弱的连接强度 strength = int(similarity * 10) + if is_cross_group: + strength = int(similarity * 5) # 降低跨群聊连接的强度 + logger.info(f"连接相似节点: {topic} 和 {similar_topic} (强度: {strength})") self.memory_graph.G.add_edge(topic, similar_topic, strength=strength, @@ -710,11 +730,12 @@ class Hippocampus: prompt = f'这是一段文字，{time_info}：{text}。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，可以包含时间和人物，以及具体的观点。只输出这句话就好' return prompt - async def _identify_topics(self, text: str) -> list: + async def _identify_topics(self, text: str, group_id: str = None) -> list: """从文本中识别可能的主题 Args: text: 输入文本 + group_id: 群聊ID，用于生成群聊特定的主题名 Returns: list: 识别出的主题列表 @@ -725,6 +746,10 @@ class Hippocampus: topics_response[0].replace("，", ",").replace("、", ",").replace(" ", ",").split(",") if topic.strip()] # print(f"话题: {topics}") + # 如果提供了群聊ID，为每个主题添加群聊标识 + if group_id: + topics = [f"{topic}_g{group_id}" for topic in topics] + return topics def _find_similar_topics(self, topics: list, similarity_threshold: float = 0.4, debug_info: str = "") -> list: @@ -747,11 +772,36 @@ class Hippocampus: # print(f"\033[1;32m[{debug_info}]\033[0m 正在思考有没有见过: {topic}") pass - topic_vector = text_to_vector(topic) + # 提取基础主题（去除群聊标识） + base_topic = topic + if "_g" in topic: + base_topic = topic.split("_g")[0] + + # 获取该主题所属的群聊ID（如果有） + topic_group_id = None + if "_g" in topic: + try: + topic_group_id = topic.split("_g")[1] + except: + pass + + topic_vector = text_to_vector(base_topic) has_similar_topic = False for memory_topic in all_memory_topics: - memory_vector = text_to_vector(memory_topic) + # 提取记忆主题的基础主题和群聊ID + base_memory_topic = memory_topic + memory_topic_group_id = None + + if "_g" in memory_topic: + parts = memory_topic.split("_g") + base_memory_topic = parts[0] + try: + memory_topic_group_id = parts[1] + except: + pass + + memory_vector = text_to_vector(base_memory_topic) # 获取所有唯一词 all_words = set(topic_vector.keys()) | set(memory_vector.keys()) # 构建向量 @@ -760,7 +810,12 @@ class Hippocampus: # 计算相似度 similarity = cosine_similarity(v1, v2) + # 如果基础主题相似 if similarity >= similarity_threshold: + # 如果两个主题来自同一群聊，提高相似度 + if topic_group_id and memory_topic_group_id and topic_group_id == memory_topic_group_id: + similarity *= 1.2 # 提高20%的相似度 + has_similar_topic = True if debug_info: # print(f"\033[1;32m[{debug_info}]\033[0m 找到相似主题: {topic} -> {memory_topic} (相似度: {similarity:.2f})") @@ -870,9 +925,17 @@ class Hippocampus: async def get_relevant_memories(self, text: str, max_topics: int = 5, similarity_threshold: float = 0.4, max_memory_num: int = 5, group_id: str = None) -> list: - """根据输入文本获取相关的记忆内容""" - # 识别主题 - identified_topics = await self._identify_topics(text) + """根据输入文本获取相关的记忆内容 + + Args: + text: 输入文本 + max_topics: 最大主题数量 + similarity_threshold: 相似度阈值 + max_memory_num: 最大记忆数量 + group_id: 群聊ID，用于优先获取指定群聊的记忆 + """ + # 识别主题，传入群聊ID + identified_topics = await self._identify_topics(text, group_id) # 查找相似主题 all_similar_topics = self._find_similar_topics( @@ -885,44 +948,58 @@ class Hippocampus: relevant_topics = self._get_top_topics(all_similar_topics, max_topics) # 获取相关记忆内容 - relevant_memories = [] + current_group_memories = [] # 当前群聊的记忆 + other_group_memories = [] # 其他群聊的记忆 + for topic, score in relevant_topics: + # 检查主题是否属于当前群聊 + topic_group_id = None + if "_g" in topic: + try: + topic_group_id = topic.split("_g")[1] + except: + pass + # 获取该主题的记忆内容 first_layer, _ = self.memory_graph.get_related_item(topic, depth=1) if first_layer: - # 如果记忆条数超过限制，随机选择指定数量的记忆 - if len(first_layer) > max_memory_num / 2: - first_layer = random.sample(first_layer, max_memory_num // 2) - - # 为每条记忆添加来源主题和相似度信息 + # 构建记忆项 for memory in first_layer: - # 添加群聊ID筛选 - # 如果memory是字典格式且有群组信息，则进行过滤 - if isinstance(memory, dict) and 'group_id' in memory: - # 当前没有指定群聊ID或者记忆来自相同群聊时才添加 - if group_id is None or memory['group_id'] == group_id: - relevant_memories.append({ - 'topic': topic, - 'similarity': score, - 'content': memory['content'] if 'content' in memory else memory, - 'group_id': memory.get('group_id') - }) + memory_item = { + 'topic': topic, + 'similarity': score, + 'content': memory if not isinstance(memory, dict) else memory.get('content', memory), + 'group_id': topic_group_id + } + + # 分类记忆 + if topic_group_id == group_id: + current_group_memories.append(memory_item) else: - # 对于没有群组信息的旧记忆，保持向后兼容 - relevant_memories.append({ - 'topic': topic, - 'similarity': score, - 'content': memory, - 'group_id': None # 旧数据没有群组信息 - }) - + other_group_memories.append(memory_item) + # 按相似度排序 - relevant_memories.sort(key=lambda x: x['similarity'], reverse=True) - - if len(relevant_memories) > max_memory_num: - relevant_memories = random.sample(relevant_memories, max_memory_num) - - return relevant_memories + current_group_memories.sort(key=lambda x: x['similarity'], reverse=True) + other_group_memories.sort(key=lambda x: x['similarity'], reverse=True) + + # 控制返回的记忆数量 + # 优先添加当前群聊的记忆，如果不足，再添加其他群聊的记忆 + final_memories = current_group_memories + remaining_slots = max_memory_num - len(final_memories) + + if remaining_slots > 0 and other_group_memories: + # 添加其他群聊的记忆，但最多添加remaining_slots个 + final_memories.extend(other_group_memories[:remaining_slots]) + + # 如果记忆总数仍然超过限制，随机采样 + if len(final_memories) > max_memory_num: + final_memories = random.sample(final_memories, max_memory_num) + + # 记录日志，显示当前群聊和其他群聊分别找到了多少记忆 + logger.debug(f"[记忆检索] 当前群聊({group_id})找到 {len(current_group_memories)} 条记忆，其他群聊找到 {len(other_group_memories)} 条记忆") + logger.debug(f"[记忆检索] 最终返回 {len(final_memories)} 条记忆") + + return final_memories def get_group_memories(self, group_id: str) -> list: """获取特定群聊的所有记忆 @@ -937,17 +1014,17 @@ class Hippocampus: all_nodes = list(self.memory_graph.G.nodes(data=True)) for concept, data in all_nodes: - memory_items = data.get('memory_items', []) - if not isinstance(memory_items, list): - memory_items = [memory_items] if memory_items else [] - - # 筛选出属于指定群聊的记忆 - for memory in memory_items: - if isinstance(memory, dict) and 'group_id' in memory and memory['group_id'] == group_id: - # 添加到结果列表 + # 检查主题是否属于指定群聊 + if f"_g{group_id}" in concept: + memory_items = data.get('memory_items', []) + if not isinstance(memory_items, list): + memory_items = [memory_items] if memory_items else [] + + # 添加所有记忆项 + for memory in memory_items: all_memories.append({ 'topic': concept, - 'content': memory.get('content', str(memory)) + 'content': memory if not isinstance(memory, dict) else memory.get('content', str(memory)) }) return all_memories