FrontierLab/ChMap-Data

ChMapData（中文记忆感知主动对话数据集）是一个新颖的数据集，专注于基于对话历史训练和评估模型在主动引入话题方面的能力，支持论文中提出的记忆感知主动对话框架。数据集包含四个关键组成部分：1）整体对话回顾，用于端到端评估；2）回调对话，用于训练记忆感知主动响应生成模型；3）对话数据，用于训练/评估话题摘要模型；4）话题排名，用于训练/评估话题检索模型。该数据集是首个关注记忆感知主动对话的中文数据集，包含训练组件和评估基准，支持对提出框架中不同模型组件的模块化评估，并提供端到端的评估协议用于全面系统评估。

The ChMapData (Chinese Memory-aware Proactive Dataset) is a novel dataset focusing on training and evaluating models capabilities in proactive topic introduction based on conversational history, supporting the memory-aware proactive dialogue framework proposed in the paper. The dataset consists of four key components: 1) Overall_dialogue_review, for end-to-end evaluation; 2) Callback_Dialogue, for training Memory-Aware Proactive Response Generation models; 3) Dialogue_Data, for training/evaluating Topic Summarization models; 4) Topic_Rank, for training/evaluating Topic Retrieval models. This dataset is the first Chinese dataset focusing on memory-aware proactive dialogue, containing both training components and evaluation benchmarks, supporting modular evaluation of different model components in the proposed framework, and providing an end-to-end evaluation protocol for comprehensive system assessment.