TOPDIAL
收藏资源简介:
TOPDIAL是由香港理工大学计算机系构建的大规模个性化目标导向对话数据集,包含约18,009条多轮对话。该数据集通过角色扮演方法自动构建,旨在解决目标导向对话系统中的个性化问题。数据集内容涵盖电影、音乐、食物和兴趣点等多个领域,通过模拟用户个人资料和性格,使系统能更自然地引导对话达到预定目标。TOPDIAL的创建过程利用了大型语言模型模拟人类社交行为,通过系统、用户和调解者三个角色的互动生成对话。该数据集主要应用于个性化目标导向对话系统的研究和开发,以提高用户体验和对话的自然性。
TOPDIAL is a large-scale personalized goal-oriented dialogue dataset constructed by the Department of Computer Science of The Hong Kong Polytechnic University, containing approximately 18,009 multi-turn dialogues. Built via automated role-playing approaches, this dataset is designed to address the personalization challenge in goal-oriented dialogue systems. Covering multiple domains including movies, music, food, and points of interest (POIs), the dataset simulates user profiles and personalities to enable systems to guide dialogues more naturally towards predetermined objectives. The construction of TOPDIAL leverages large language models to simulate human social behaviors, generating dialogues through interactions among three roles: the system, user, and mediator. This dataset is primarily applied to the research and development of personalized goal-oriented dialogue systems, aiming to enhance user experience and the naturalness of dialogues.




