Improved graph convolutional network for emotion analysis in social media text

Understanding emotions in social media text is crucial for mental health monitoring, sentiment analysis, and misinformation detection applications. This study presents an Improved Graph Convolutional Network (IGCN) that leverages the network structure of social media text to enhance emotion classification. Unlike conventional GCN models, our approach integrates a Pointwise Mutual Information (PMI) based graph construction method, which improves the representation of semantic relationships between words. Additionally, an attention mechanism selectively emphasizes contextually significant words, enhancing interpretability and classification accuracy. <b>Key contributions:</b> •Graph-based sentiment modelling to capture deep semantic relationships in text.•Improved interpretability through attention-weighted word importance.•Scalability for large social media datasets, ensuring efficient processing. Through extensive comparative experiments, our model achieves 78.64% and 92.38% accuracy, demonstrating the effectiveness of GNNs in large-scale emotion classification. This research underscores the transformative potential of graph-based NLP models for decoding emotional tones in social media, paving the way for more accurate and insightful sentiment analysis. The research is conducted on two large-scale datasets: Twitter_EA: Categorizing tweets into six emotions-Sadness, Joy, Love, Anger, Fear, and Surprise. Emotion Recognition Dataset: Labelling emotions as Anxiety, Depression, Happiness, and Stress.