EmoDynamiX: Emotional Support Dialogue Strategy Prediction by Modelling MiXed Emotions and Discourse Dynamics

2024-08-16

arXiv.org

AI Summary

To address the opacity in strategy selection and severe model preference bias in emotional support conversations (ESC), this paper proposes a novel decoupled framework that separates strategy prediction from language generation. Methodologically, it introduces the first heterogeneous graph-based approach for dynamic emotion–strategy modeling, explicitly capturing discourse-level evolutionary relationships between users’ fine-grained emotional states and system response strategies. The method integrates heterogeneous graph neural networks, fine-grained emotion representation, and discourse-aware dynamic modeling to ensure traceable strategy decisions, reduced preference bias, and enhanced interpretability. Evaluated on two standard ESC benchmarks, our approach achieves significant improvements over state-of-the-art methods: it substantially enhances strategy proficiency while reducing average preference bias by 23.6%. Moreover, it enables end-to-end backward tracing and attribution analysis of the decision process, thereby improving transparency and accountability in emotional support generation.

Technology Category

• Natural Language Processing: Discourse, Pragmatics & Argument Mining
• Humans and AI: Emotional Intelligence
• Knowledge Representation and Reasoning: Preferences

Application Category

• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization
• 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: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Designing emotionally intelligent conversational systems to provide comfort and advice to people experiencing distress is a compelling area of research. Recently, with advancements in large language models (LLMs), end-to-end dialogue agents without explicit strategy prediction steps have become prevalent. However, implicit strategy planning lacks transparency, and recent studies show that LLMs' inherent preference bias towards certain socio-emotional strategies hinders the delivery of high-quality emotional support. To address this challenge, we propose decoupling strategy prediction from language generation, and introduce a novel dialogue strategy prediction framework, EmoDynamiX, which models the discourse dynamics between user fine-grained emotions and system strategies using a heterogeneous graph for better performance and transparency. Experimental results on two ESC datasets show EmoDynamiX outperforms previous state-of-the-art methods with a significant margin (better proficiency and lower preference bias). Our approach also exhibits better transparency by allowing backtracing of decision making.

Problem

• Emotional Support
• Complex Emotions Understanding
• Transparency in Decision-making

Innovation

• EmoDynamiX
• Graph Analysis
• Bias Reduction