CrossWOZ: A Large-Scale Chinese Cross-Domain Task-Oriented Dialogue Dataset

To advance multi-domain (cross-domain) dialogue modeling as well as alleviate\nthe shortage of Chinese task-oriented datasets, we propose CrossWOZ, the first\nlarge-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. It\ncontains 6K dialogue sessions and 102K utterances for 5 domains, including\nhotel, restaurant, attraction, metro, and taxi. Moreover, the corpus contains\nrich annotation of dialogue states and dialogue acts at both user and system\nsides. About 60% of the dialogues have cross-domain user goals that favor\ninter-domain dependency and encourage natural transition across domains in\nconversation. We also provide a user simulator and several benchmark models for\npipelined task-oriented dialogue systems, which will facilitate researchers to\ncompare and evaluate their models on this corpus. The large size and rich\nannotation of CrossWOZ make it suitable to investigate a variety of tasks in\ncross-domain dialogue modeling, such as dialogue state tracking, policy\nlearning, user simulation, etc.\n