Relational Temporal Graph Reasoning for Dual-Task Dialogue Language Understanding.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Machine Learning

Background:

• Dual-task dialog language understanding leverages correlations between related tasks.
• Existing methods often rely on implicit semantic interactions.

Purpose of the Study:

• To propose a novel framework for dual-task dialog language understanding based on relational temporal graph reasoning.
• To enhance interaction modeling by incorporating explicit prediction-level dependencies.

Main Methods:

• Developed a speaker-aware temporal graph (SATG) and a dual-task relational temporal graph (DRTG).
• Introduced the Dual-tAsk temporal Relational rEcurrent Reasoning network (DARER) using relational graph convolutional networks (RGCNs).
• Proposed the Relational Temporal Transformer (ReTeFormer) for fine-grained relational temporal modeling, leading to DARER².

Main Results:

• DARER and DARER² demonstrated superior performance over state-of-the-art models.
• Achieved significant relative improvements of 28% (DARER) and 34% (DARER²) in F1 score for dialog sentiment classification on the Mastodon dataset.

Conclusions:

• The proposed relational temporal graph reasoning framework effectively models dual-task dialog language understanding.
• Explicit prediction-level interactions and advanced temporal modeling enhance model performance.