HYU-NLP/TACT

TACT（TOD-And-Chitchat Transition）是一个用于过渡感知对话建模的数据集，旨在解决任务导向对话(TOD)和闲聊对话之间的自然过渡问题。该数据集支持用户驱动和代理驱动的模式切换，包含两个部分：基于MultiWOZ的TACT_multiwoz和基于SLURP的TACT_slurp。数据集可用于训练和评估对话系统在意图识别和模式转换处理方面的能力。通过结合结构多样的数据和直接偏好优化(DPO)方法，该数据集有助于开发更具主动性和过渡感知能力的对话代理。

TACT (TOD-And-Chitchat Transition) is a dataset designed for transition-aware dialogue modeling that addresses the natural transitions between task-oriented dialogue (TOD) and open-ended chitchat. The dataset supports both user- and agent-driven mode switches and consists of two parts: TACT_multiwoz (based on MultiWOZ) and TACT_slurp (based on SLURP). It can be used to train and evaluate dialogue systems capabilities in intent detection and mode transition handling. By combining structurally diverse data with Direct Preference Optimization (DPO), the dataset facilitates the development of more proactive and transition-aware conversational agents.