Abstract

While large language model (LLM)-based chatbots have been applied foreffective engagement in credit dialogues, their capacity for dynamic emotionalexpression remains limited. Current agents primarily rely on passive empathyrather than affective reasoning. For instance, when faced with persistentclient negativity, the agent should employ strategic emotional adaptation byexpressing measured anger to discourage counterproductive behavior and guidethe conversation toward resolution. This context-aware emotional modulation isessential for imitating the nuanced decision-making of human negotiators. Thispaper introduces an EQ-negotiator that combines emotion sensing frompre-trained language models (PLMs) with emotional reasoning based on GameTheory and Hidden Markov Models. It takes into account both the current andhistorical emotions of the client to better manage and address negativeemotions during interactions. By fine-tuning pre-trained language models (PLMs)on public emotion datasets and validating them on the credit dialogue datasets,our approach enables LLM-based agents to effectively capture shifts in clientemotions and dynamically adjust their response tone based on our emotiondecision policies in real-world financial negotiations. This EQ-negotiator canalso help credit agencies foster positive client relationships, enhancingsatisfaction in credit services.