Emotionally Intelligent Task-oriented Dialogue Systems: Architecture, Representation, and Optimisation

2025-07-02

AI Summary

Task-oriented dialogue systems struggle to simultaneously achieve high task success rates, accurate emotion understanding, and information fidelity in noisy and ambiguous environments. To address this, we propose LUSTER—a unified framework integrating large language models (LLMs) with end-to-end reinforcement learning. LUSTER introduces a novel joint optimization mechanism that concurrently maximizes short-term emotional reward and long-term task reward, and incorporates a natural language understanding module alongside a simulator-based user behavior model. Under structured reward guidance, LUSTER significantly improves emotion recognition accuracy (+12.3%) and task completion rate (+9.8%) over baseline methods. It further enhances dialogue robustness and user experience. By unifying emotion-aware reasoning with task-driven policy learning in an end-to-end trainable architecture, LUSTER establishes a scalable, optimization-centric paradigm for affective intelligent task-oriented dialogue systems.

Technology Category

• Natural Language Processing: Conversational AI/Dialog Systems
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Task-oriented dialogue (ToD) systems are designed to help users achieve specific goals through natural language interaction. While recent advances in large language models (LLMs) have significantly improved linguistic fluency and contextual understanding, building effective and emotionally intelligent ToD systems remains a complex challenge. Effective ToD systems must optimise for task success, emotional understanding and responsiveness, and precise information conveyance, all within inherently noisy and ambiguous conversational environments. In this work, we investigate architectural, representational, optimisational as well as emotional considerations of ToD systems. We set up systems covering these design considerations with a challenging evaluation environment composed of a natural-language user simulator coupled with an imperfect natural language understanding module. We propose extbf{LUSTER}, an extbf{L}LM-based extbf{U}nified extbf{S}ystem for extbf{T}ask-oriented dialogue with extbf{E}nd-to-end extbf{R}einforcement learning with both short-term (user sentiment) and long-term (task success) rewards. Our findings demonstrate that combining LLM capability with structured reward modelling leads to more resilient and emotionally responsive ToD systems, offering a practical path forward for next-generation conversational agents.

Problem

• Build emotionally intelligent task-oriented dialogue systems
• Optimize task success and emotional responsiveness in noisy environments
• Combine LLM capability with structured reward modeling

Innovation

• LLM-based unified system for task-oriented dialogue
• End-to-end reinforcement learning with dual rewards
• Combining LLM capability with structured reward modeling