FiSMiness: A Finite State Machine Based Paradigm for Emotional Support Conversations

TL;DR

FiSMiness introduces a Finite State Machine–driven paradigm for emotional support conversations, enabling an LLM to bootstrap planning by jointly inferring seeker's emotion, selecting a support strategy, and generating a response at each turn. The framework defines a 5-tuple FSM with states $s_0$–$s_3$ and transitions that cycle via $e(t)$, $g(t)$, and $u^p(t)$, with finetuning conducted on the ESConv dataset to align multi-task objectives. Empirical results across automatic metrics, human judgments, and LLM-based evaluation show FiSMiness and its variants outperform direct-inference, prompting-based, fine-tuning, and external-knowledge baselines, especially in long-horizon turns. The work demonstrates the benefit of structuring ESC as an FSM-guided, multi-step reasoning process and discusses limitations such as simplistic transitions and single-strategy outputs, outlining potential future integration with reinforcement learning and combined paradigms for richer interactions.

Abstract

Emotional support conversation (ESC) aims to alleviate the emotional distress of individuals through effective conversations. Although large language models (LLMs) have obtained remarkable progress on ESC, most of these studies might not define the diagram from the state model perspective, therefore providing a suboptimal solution for long-term satisfaction. To address such an issue, we leverage the Finite State Machine (FSM) on LLMs, and propose a framework called FiSMiness. Our framework allows a single LLM to bootstrap the planning during ESC, and self-reason the seeker's emotion, support strategy and the final response upon each conversational turn. Substantial experiments on ESC datasets suggest that FiSMiness outperforms many baselines, including direct inference, self-refine, chain of thought, finetuning, and external-assisted methods, even those with many more parameters.

Figures (3)

• Figure 1: The paradigm of FiSMiness.
• Figure 2: The FiSMiness framework.
• Figure 3: Rouge-L (R-L) results as function of conversation turns. Results are collected from ESconv.