Emotion inference in conversations based on commonsense enhancement and graph structures.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Dialogue emotion inference often lacks common sense knowledge, hindering accuracy.
• Traditional methods struggle to extract dialogue's structural and semantic features effectively.
• Existing models face limitations in understanding complex dialogue structures and integrating external knowledge.

Purpose of the Study:

• To propose a novel dialogue emotion inference model named CEICG.
• To address the limitations of current models by integrating common sense knowledge.
• To enhance the accuracy of emotion inference in conversational contexts.

Main Methods:

• Developed a dialogue emotion inference model based on Common Sense Enhancement and Graph Model (CEICG).
• Dynamically constructed graph nodes and defined diverse edge relations to simulate dialogue evolution.
• Integrated external common sense knowledge into the graph model using two distinct methods.

Main Results:

• The CEICG model demonstrated superior performance in emotion inference tasks.
• Outperformed six existing baseline models across three different datasets.
• Effectively captured structural and semantic features of conversations by simulating dialogue evolution.

Conclusions:

• Integrating external common sense knowledge significantly enhances dialogue emotion inference capabilities.
• The CEICG model offers a more robust approach to understanding emotions in dialogue.
• The proposed methods overcome previous limitations in processing complex dialogue structures and knowledge integration.