A Computational Approach to Understanding Empathy Expressed in Text-Based Mental Health Support

摘要: Empathy is critical to successful mental health support. measurement has predominantly occurred in synchronous, face-to-face settings, and may not translate asynchronous, text-based contexts. Because millions of people use platforms for support, understanding empathy these contexts crucial. In this work, we present a computational approach how expressed online platforms. We develop novel unifying theoretically-grounded framework characterizing the communication conversations. collect share corpus 10k (post, response) pairs annotated using with supporting evidence annotations (rationales). multi-task RoBERTa-based bi-encoder model identifying conversations extracting rationales underlying its predictions. Experiments demonstrate that our can effectively identify empathic further apply analyze 235k interactions show users do self-learn over time, revealing opportunities training feedback.