TherapyTalk

This dataset consists of 104 Reddit-based conversational examples and is designed to serve as a test set for tasks involving the generation of emotionally supportive responses. Each instance includes three sequential posts (post1, post2, and post3) that provide the context of a conversation, along with a corresponding response written by a human annotator. The raw field contains the original Reddit thread data, including metadata such as the author, timestamp, content of each post, and the subreddit it was posted in. This structure allows models to understand the progression of a conversation and generate context-aware, empathetic responses. The dataset is in English, licensed under CC BY 4.0, and is suitable for evaluation in emotional support dialogue systems or other applications requiring sensitive and contextually appropriate replies.