ENPMR-Bench
收藏资源简介:
ENPMR-Bench是由哈尔滨工业大学社会计算与人机交互研究中心构建的中文基准数据集,旨在系统评估情感需求感知的主动记忆检索能力。该数据集包含1,872个记忆增强的情感支持对话,覆盖11,846条记忆条目,涵盖生理需求、爱与归属、自尊及自我实现四大需求维度,数据来源于基于用户画像和生命主题生成的合成对话。其构建过程以马斯洛需求层次理论为指导,通过专家设计的结构化检索框架,将记忆条目分类为高光时刻、能力、关系、目标和偏好五类,并按照时间线组织成连贯的会话历史。该数据集主要应用于情感计算和对话系统领域,旨在解决长期情感支持场景中,智能体如何主动推断用户潜在情感需求并检索恰当记忆以提供共情交互的核心问题。
ENPMR-Bench is a Chinese benchmark dataset constructed by the Research Center for Social Computing and Human-Computer Interaction of Harbin Institute of Technology, which aims to systematically evaluate the active memory retrieval capability for emotional demand perception. This dataset contains 1,872 memory-augmented emotional support conversations, covering 11,846 memory entries, and spans four major demand dimensions: physiological needs, love and belonging, self-esteem, and self-actualization. The data is derived from synthetic conversations generated based on user profiles and life themes. Its construction is guided by Maslow's Hierarchy of Needs. Through a structured retrieval framework designed by experts, memory entries are categorized into five types: highlight moments, abilities, relationships, goals, and preferences, and organized into coherent conversational histories along a timeline. This dataset is primarily applied in the fields of affective computing and conversational systems, aiming to address the core problem of how AI agents actively infer users' potential emotional needs and retrieve appropriate memories to provide empathetic interactions in long-term emotional support scenarios.



