EmoFSM: A Finite State Machine for Emotional Support Conversation

The article introduces EmoFSM, a framework that integrates Finite State Machines (FSM) with large language models (LLMs) to enhance emotional support conversations (ESC). EmoFSM enables the model to plan and adapt responses based on the seeker's emotions and support strategies during each turn of conversation. Experimental results indicate that EmoFSM significantly outperforms various baseline methods, including direct inference and self-fine-tuning, regardless of parameter size, highlighting its potential for improving long-term user satisfaction in ESC applications.