HYBRIDIALOGUE

HYBRIDIALOGUE是一个创新的信息寻求对话数据集，由加利福尼亚理工学院等机构创建。该数据集包含超过4800个样本，涉及自然对话，这些对话基于维基百科的文本和表格数据。数据集通过分解复杂的多跳问题为简单的多轮对话交互来创建，旨在解决对话系统中跨不同模态信息分布的挑战。HYBRIDIALOGUE不仅用于训练对话系统，还支持包括检索、系统状态跟踪和对话响应生成等多项任务，展示了在处理复杂信息寻求对话场景中的应用潜力。

HYBRIDIALOGUE is an innovative information-seeking dialogue dataset created by institutions such as the California Institute of Technology. It contains over 4,800 natural dialogue samples grounded in Wikipedia text and tabular data. The dataset was constructed by decomposing complex multi-hop questions into simple multi-turn dialogue interactions, with the aim of addressing the challenges of cross-modal information distribution in dialogue systems. HYBRIDIALOGUE supports not only the training of dialogue systems, but also a variety of tasks including retrieval, system state tracking, and dialogue response generation, showcasing its application potential in handling complex information-seeking dialogue scenarios.