GEM: Graph-Enhanced Mixture-of-Experts with ReAct Agents for Dialogue State Tracking

TL;DR

GEM introduces a graph-enhanced mixture-of-experts framework with ReAct agents, significantly improving dialogue state tracking accuracy by combining structured dialogue understanding, dynamic expert routing, and agent-based reasoning.

Contribution

The paper presents a novel architecture that integrates graph neural networks, mixture-of-experts, and ReAct agents for enhanced dialogue state tracking performance.

Findings

GEM achieves 65.19% Joint Goal Accuracy on MultiWOZ 2.2.

Outperforms existing end-to-end LLM approaches and SOTA methods.

Demonstrates the effectiveness of combining structured dialogue representation with expert routing and reasoning.

Abstract

Dialogue State Tracking (DST) requires precise extraction of structured information from multi-domain conversations, a task where Large Language Models (LLMs) struggle despite their impressive general capabilities. We present GEM (Graph-Enhanced Mixture-of-Experts), a novel framework that combines language models and graph-structured dialogue understanding with ReAct agent-based reasoning for superior DST performance. Our approach dynamically routes between specialized experts: a Graph Neural Network that captures dialogue structure and turn-level dependencies, and a finetuned T5-Small encoder-decoder for sequence modeling, coordinated by an intelligent router. For complex value generation tasks, we integrate ReAct agents that perform structured reasoning over dialogue context. On MultiWOZ 2.2, GEM achieves 65.19% Joint Goal Accuracy, substantially outperforming end-to-end LLM approaches (best: 38.43%) and surpassing state-of-the-art (SOTA) methods including TOATOD (63.79%), D3ST (58.70%), and Diable (56.48%). Our graph-enhanced mixture-of-experts architecture with ReAct integration demonstrates that combining structured dialogue representation with dynamic expert routing and agent-based reasoning provides a powerful paradigm for dialogue state tracking, achieving superior accuracy while maintaining computational efficiency through selective expert activation.