Integrating Embedding Representations with Graph Convolutional Networks for Enhanced Sentiment Analysis

This paper explores the application of Graph Convolutional Networks (GCNs) for sentiment analysis in text classification by leveraging graph-based text representations. To address the limitations of traditional text classification, we constructed a graph where nodes represent words and documents, and edges capture syntactic and semantic relationships based on co-occurrence and dependency structures within the text. We utilize Word2Vec embeddings to initialize node features, transforming the text graph into a format for GCN-based sentiment classification. The proposed pipeline was evaluated on the IMDB movie review dataset, achieving a classification accuracy of 90.64% and outperformed other state-of-the-art methods such as recurrent neural networks, convolutional neural networks, and transformer-based models applied to this task. This work demonstrates that GCNs, when applied to text graphs, effectively capture relational information in sentiment classification, providing an alternative approach to more commonly used sequential or attention-based models.