RuiSumida/LUFY

LUFY是一个长对话数据集，旨在研究检索增强生成（RAG）聊天机器人中的选择性遗忘和长期记忆管理。该数据集包含人类用户与AI助手之间的自然对话，并附有结构化的问答对和证据注释，明确将答案与对话轮次关联起来。这使得研究人员能够研究对话代理中的记忆选择、遗忘、检索和事实一致性。数据集分为两种配置：turns（对话轮次）和qa（问答对）。turns配置包含每个对话轮次的详细信息，如用户ID、对话ID、角色和内容等；qa配置则包含从对话中提取的问答对及其支持的证据轮次ID。数据集还提供了详细的统计信息，如用户数量、对话长度和轮次数量等。

LUFY is a long-form conversational dataset designed to study selective forgetting and long-term memory management in Retrieval-Augmented Generation (RAG) chatbots. The dataset contains extended, natural conversations between human users and an AI assistant, enriched with structured question–answer (QA) pairs and evidence annotations that explicitly ground answers in dialogue turns. This enables research on memory selection, forgetting, retrieval, and factual consistency in conversational agents. The dataset is released in two configurations: turns (dialogue turns) and qa (question-answer pairs). The turns configuration includes details of each dialogue turn, such as user ID, conversation ID, role, and content. The qa configuration contains question-answer pairs derived from conversations and their supporting evidence turn IDs. The dataset also provides detailed statistics, such as the number of users, conversation length, and turn count.