Exploring the potential of graph neural networks for Vietnamese sentiment analysis

Abstract

Graph Neural Networks (GNNs) have demonstrated outstanding potential in various natural language processing (NLP) tasks; however, their application to Vietnamese sentiment analysis remains relatively underexplored. This study evaluates the effectiveness of several GNN-based models – including TextGCN, HGAT, BertGCN, and GraphSAGE – in analyzing sentiments in Vietnamese textual data. Experiments were conducted on two benchmark Vietnamese sentiment datasets, UIT-VSFC and Foody. The empirical results compare the performance of GNN models with both traditional machine learning approaches and representative deep learning architectures. Key performance metrics such as accuracy and F1-score were analyzed to highlight the strengths of each method. The findings reveal that GNN-based models exhibit superior capabilities in capturing contextual and semantic relationships within texts, particularly in complex sentiment scenarios. This study aims to investigate the potential of applying GNNs to enhance Vietnamese sentiment analysis, offering a novel perspective compared to traditional and deep learning models. Additionally, the implementation code of the GNN models has been made available on GitHub to serve as a resource for other research groups interested in this domain.