记忆模块

联想记忆

定义生成式智能体的核心长期记忆模块
源代码相对路径:reverie\backend_server\persona\memory_structures\associative_memory.py

ConceptNode 概念节点

class ConceptNode: 
  """
  概念节点类 - 记忆流中的基本单元
  
  每个 ConceptNode 代表智能体记忆中的一个概念，可以是：
  - 事件(event): 观察到的行为或状态变化，深度=0
  - 思考(thought): 基于事件的反思和抽象，深度≥1  
  - 对话(chat): 与其他智能体的交流记录，深度=0
  """
  def __init__(self,
               node_id, node_count, type_count, node_type, depth,
               created, expiration, 
               s, p, o, 
               description, embedding_key, poignancy, keywords, filling): 
    # === 节点标识信息 ===
    self.node_id = node_id           # 唯一ID: "node_1", "node_2"...
    self.node_count = node_count     # 全局节点计数器
    self.type_count = type_count     # 同类型节点计数器
    self.type = node_type           # 节点类型: "event"/"thought"/"chat"
    self.depth = depth              # 抽象层级: 0=原始观察, 1+=反思层级

    # === 时间属性 ===
    self.created = created          # 创建时间(datetime对象)
    self.expiration = expiration    # 过期时间(可选，用于临时记忆)
    self.last_accessed = self.created  # 最后访问时间(用于时近性计算)

    # === SPO三元组结构 (主语-谓语-宾语) ===
    self.subject = s                # 主语: 谁/什么
    self.predicate = p              # 谓语: 做什么/是什么状态
    self.object = o                 # 宾语: 对什么/在哪里

    # === 内容和元数据 ===
    self.description = description   # 人类可读的描述文本
    self.embedding_key = embedding_key  # 向量嵌入的键名
    self.poignancy = poignancy      # 重要性/情感强度评分(1-10)
    self.keywords = keywords        # 关键词集合(用于快速检索)
    self.filling = filling          # 填充信息(用于思考节点的支撑证据)

  def spo_summary(self): 
    """
    返回SPO三元组摘要
    
    返回:
        tuple: (主语, 谓语, 宾语) 的元组形式
    
    示例:
        ("Isabella Rodriguez", "is", "sleeping") 
        ("咖啡馆", "位于", "小镇中心")
    """
    return (self.subject, self.predicate, self.object)

节点是这样的数据结构 如nodes.json

{
    "node_87": {
        "node_count": 87,
        "type_count": 81,
        "type": "event",
        "depth": 0,
        "created": "2023-02-13 09:00:20",
        "expiration": null,
        "subject": "the Ville:Oak Hill College:library:library table",
        "predicate": "is",
        "object": "use",
        "description": "library table is in use",
        "embedding_key": "library table is in use",
        "poignancy": 3,
        "keywords": [
            "library table",
            "use"
        ],
        "filling": []
    },
    "node_86": {
        "node_count": 86,
        "type_count": 80,
        "type": "event",
        "depth": 0,
        "created": "2023-02-13 09:00:10",
        "expiration": null,
        "subject": "Klaus Mueller",
        "predicate": "research",
        "object": "topic",
        "description": "Klaus Mueller is researching the topic",
        "embedding_key": "researching the topic",
        "poignancy": 8,
        "keywords": [
            "topic",
            "Klaus Mueller"
        ],
        "filling": []
    },
}