Emotionally intelligent preference optimization for conversational behavior change support

Abstract

Procrastination negatively affects productivity and well-being, yet current large language models (LLMs) often lack the nuanced emotional understanding necessary for effective behavior-change support. This study proposes emotionally intelligent preference optimization (EIPO), a unified framework integrating emotional chain-of-thought reasoning, teacher-guided self-correction, and direct preference optimization to enhance emotional intelligence in conversational user interfaces. Through ablation experiments evaluated using the emotional generation score (EGS), we demonstrate that EIPO significantly improves empathetic and constructive responses in procrastination-related interactions. Crucially, cross-model validation on Qwen2.5-7B, Llama-3.2-3B, and Gemma-2-2B confirms EIPO’s model-agnostic design, with consistent EGS gains ( $$\Delta = +0.13$$ to $$+0.27$$ ) and robust independent evaluation using Claude-3.5-Sonnet. These findings establish EIPO as a principled, transferable framework for embedding emotionally intelligent behavior-change support into diverse LLM-driven CUIs.