CKG: Improving ABSA with text augmentation using ChatGPT and knowledge-enhanced gated attention graph convolutional...

Area of Science:

• Natural Language Processing
• Neurorobotics

Background:

• Aspect-level sentiment analysis (ABSA) is crucial for understanding nuanced emotions in text.
• Limited training data and feature integration issues hinder current ABSA models.
• Previous methods using graph convolutions face gradient explosion problems.

Purpose of the Study:

• To investigate ChatGPT for aspect-level text augmentation.
• To introduce an improved gated attention mechanism for graph convolutional networks.
• To enhance the integration of semantic and syntactic features for ABSA.

Main Methods:

• Leveraging ChatGPT for aspect-level text augmentation.
• Developing an improved gated attention mechanism to mitigate gradient explosion in graph convolutional networks.
• Enriching dependency graph features with a sentiment knowledge base and employing cross-fusion for feature integration.

Main Results:

• The proposed model demonstrates superior performance compared to baseline models.
• Effective integration of semantic and syntactic features was achieved.
• The improved gated attention mechanism successfully addressed gradient explosion.

Conclusions:

• The study presents a novel approach to ABSA using ChatGPT and an enhanced graph convolutional network.
• The findings highlight the effectiveness of the proposed methods in overcoming data scarcity and feature integration challenges.
• The model shows significant improvements in sentiment polarity prediction accuracy.