Corpus linguistic-driven hybrid feature representation learning with graph convolutional network for text-based emotion detection

Text-based emotion detection becomes a significant research area in natural language processing (NLP), supporting various domains such as diagnosing mental health status, customer feedback analysis, and social media monitoring. However, accurate detection of emotion classes is a challenging task as human emotional states are contextual, uncertain, and subtle. Existing emotion detection models typically struggle to fully understand these underlying emotional characteristics. To address these issues, this research introduces a Hybrid Attention Learning with Graph Convolutional Network for Text-Based Emotion Detection (HALGCN-TED) framework, aiming to develop a robust and automated model to improve the representation and detection of complex emotional patterns. The model uses a hybrid feature embedding technique by combining RoBERTa and FastText, where RoBERTa captures contextual semantic connections, while FastText learns sub-word and word-level patterns effectively. These representations are further fused through an attention-based feature fusion strategy to provide highly informative features and reduce the impact of less informative features. The model applies a graph convolutional network to learn complex relationships among these fused representations and facilitate more accurate separation of multiple emotional classes. Furthermore, the model employs AdamW optimization to improve the model's learning stability and support more effective convergence. An extensive experimental analysis is performed using two benchmark datasets across various performance metrics to verify the performance of the HALGCN-TED framework in emotion detection from text datasets. These results show that the proposed model achieves excellent outcomes over compared models and validates its potential in improved learning representations and accurate classification of multiple emotional classes.