BERT-based two-channel neural network model text emotion analysis

Introduction: Text emotion analysis, or sentiment analysis, is a crucial area in natural language processing (NLP) focused on identifying emotions within textual data. Challenges in this field include disappearing gradients, information loss, and the lack of contextual semantics.Methods: To address these challenges, we propose a BERT-based model utilizing a two-channel neural network for enhanced emotion classification. The model transforms text into word vectors using BERT, which excels in capturing contextual information. The architecture includes Augmented Recurrent Neural Networks-mutated Unidirectional Long Short-Term Memory (ARNN-Uni-LSTM) to extract local semantic features and capture long-range dependencies. Preprocessing involved tokenization and Word2Vec on publicly available text emotion datasets. The first channel employs ARNN for local feature extraction, while the second uses Uni-LSTM for broader context.Results: Experiments conducted in Python demonstrated that our model outperformed traditional methods, achieving precision of 97.18%, recall of 94.56%, and an F1 score of 96.26%.Conclusions: The BERT-based model shows significant promise for applications such as customer feedback analysis, social media monitoring, and mental health diagnostics, offering a foundation for advanced emotion recognition systems.