MyGovEmotion-GNN: A Graph Neural Network Framework for Emotion Recognition on MyGov Comments

In Natural language processing (NLP), Emotion identification from text is an essential task, with a wide range of applications from sentiment analysis to customer feedback systems, government schemes monitoring, and mental health monitoring. In this study, we introduce MyGovEmotion-GNN, a novel model for textual emotion classification on MyGov comments. Our Approach that uses Graph Neural Networks (GNN) to bridge the gap between traditional text representation techniques and graph-based learning. In contrast to conventional models that only use transformer-based architectures or sequential encoding, our approach makes use of both structural context and semantic richness to predict emotions more reliably. Using cosine similarity of Term Frequency-Inverse Document Frequency (TF-IDF) vectors, our methodology involves modeling text documents as graphs, with nodes representing documents and edges expressing crucial connections. To recognize emotions in comments, we have employed Graph Convolutional Networks (GCN) under GNN based framework to categorize reviews. In our work, we have utilized comments dataset collected from MyGov political website incorporating government schemes and publicly available GoEmotions dataset. Also, we have performed experimental evaluation using publicly available benchmark dataset such as GoEmotions. Further, the experimental results illustrate that the proposed GNN model achieved the better results, attaining 0.67 accuracy score and F1 score of 0.66.