Enhancing Goal-oriented Proactive Dialogue Systems via Consistency Reflection and Correction

2025-06-16

AI Summary

To address intent drift and response inconsistency in goal-oriented proactive dialogue systems, this paper proposes a consistency reflection and correction mechanism. Our method uniquely integrates dialogue state tracking (DST) with explicit goal graph modeling to construct an LLM-based consistency discriminator that dynamically detects logical conflicts among dialogue state, user goals, and system responses. A lightweight correction decoder then performs real-time response adjustment. Evaluated on MultiWOZ and SGD benchmarks, our approach improves task success rate by 8.2%, reduces inconsistency error rate by 37%, and incurs less than 5% additional inference latency. These results demonstrate substantial gains in task completion and user satisfaction, establishing a novel paradigm for controllable and interpretable proactive dialogue systems.

Technology Category

• Natural Language Processing: Conversational AI/Dialog Systems
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

This paper proposes a consistency reflection and correction method for goal-oriented dialogue systems.

Problem

• Improving goal-oriented dialogue systems' consistency
• Proposing reflection and correction methods
• Enhancing proactive dialogue system performance

Innovation

• Consistency reflection method for dialogue systems
• Correction technique for goal-oriented dialogues
• Proactive dialogue enhancement via reflection