CareCall

该数据集旨在研究长期对话中的动态记忆变化，基于现有的CareCall数据集构建而成，后者包含的是开放领域对话的单次会话。新数据集将这些单次会话扩展到多会话环境，从而实现对记忆管理和角色个性更新的处理。在生成此数据集的过程中，使用了大规模语言模型，并经过自动生成后的人工修订，以确保对话质量。该数据集包含了600个对话会话，每个会话超过15个交流回合，其任务重点在于长期对话中的记忆管理。

This dataset aims to study dynamic memory changes in long-term conversations. It is constructed based on the existing CareCall dataset, which contains single-session open-domain dialogues. The new dataset expands these single-session dialogues into multi-session settings, enabling research on memory management and persona updating. During the dataset development process, large language models (LLMs) were employed for automated generation, followed by manual revisions to ensure dialogue quality. This dataset comprises 600 dialogue sessions, each with more than 15 conversational turns, and its core task focuses on memory management in long-term conversations.