REALTALK
收藏资源简介:
REALTALK是一个真实世界的长期对话数据集,由10对个体在21天内通过消息应用进行的真实对话组成。数据集包含大约16,000个词/对话,旨在通过直接比较真实世界对话和大型语言模型生成的对话,来研究情感智能属性和角色一致性等挑战。该数据集用于评估模型在长期对话中的情感智能和角色模拟能力,并为开发更加人性化的记忆感知AI提供基准任务。
REALTALK is a real-world long-term conversation dataset consisting of authentic dialogues between 10 pairs of individuals conducted over 21 days via messaging applications. The dataset contains approximately 16,000 words across all conversations. It aims to investigate challenges including emotional intelligence attributes and role consistency through direct comparisons between real-world dialogues and those generated by large language models (LLMs). This dataset is utilized to evaluate models' emotional intelligence and role simulation capabilities in long-term conversations, and provides benchmark tasks for developing more human-like memory-aware AI.




