Graph Neural Networks for Text-Based Emotion Detection: A Comprehensive Framework for Social Media Analysis

Abstract

Social media analysis requires that one be able to interpret emotions through textual information; the classic models of NLP usually do not recognize the nuances of emotions. In this paper, the new Graph Neural Network (GNN)-based emotion recognition framework is introduced and the nodes thing is the words and the edges thing is the semantic relationships that model both local and global contextual dependencies. The suggested architecture combines Graph Convolutional Networks (GCN), Graph Attention Networks (GAT), and GraphSAGE layers to improve the feature learning and semantic representation. The evaluation on standard datasets, such as the IMDB and multi-class emotion classification tasks, had an accuracy of 92.17% and 89.3%, respectively, which was higher than the traditional deep-learning methods, including CNN, LSTM, and BERT. Moreover, the framework is both interpretable, through finding emotion-inducing word configurations, and is capable of dealing with the variable text length and long-range dependencies, which renders it very effective in emotion analysis when applied to social media environments.