A Graph Convolutional Network Methodology for Aspect-Based Sentiment Analysis Focusing on Aspect-Term Relations

Aspect-based sentiment Analysis (ABSA) aims to determine sentiment polarity concerning specific aspect terms within a sentence. Traditional Graph Convolutional Networks (GCNs) often capture all word relations, introducing noise and reducing accuracy. This paper introduces a novel GCN methodology that captures only aspect-related terms, supported by a math-ematical framework, to enhance sentiment classification accuracy. Our model leverages dependency parsing to construct aspect-centric subgraphs, demonstrating significant improvements on benchmark ABSA datasets.