IPM-PrefDial
收藏资源简介:
IPM-PrefDial数据集是浙江大学团队创建的,旨在解决情感支持对话中存在的心理策略与响应内容纠缠的问题。数据集通过推断性偏好挖掘方法构建,包含21370个策略偏好对和11887个响应偏好对。该数据集旨在为情感支持对话任务中的策略规划和响应生成提供高质量的监督信号,并用于训练去耦情感支持对话框架,以提高对话质量和同情心。
The IPM-PrefDial dataset was developed by a team from Zhejiang University to address the entanglement between psychological strategies and response content in emotional support conversations. Constructed via an inferential preference mining approach, this dataset contains 21,370 strategy-preference pairs and 11,887 response-preference pairs. It is designed to provide high-quality supervision signals for strategy planning and response generation in emotional support dialogue tasks, and can be used to train decoupled emotional support dialogue frameworks to improve dialogue quality and empathy.

- 1DecoupledESC: Enhancing Emotional Support Generation via Strategy-Response Decoupled Preference Optimization浙江大学 · 2025年



