Enhancing Emotion Classification through Neural Networks and Data Augmentation

Emotion classification is a key task in natural language processing with lots of applications in our daily life.Traditional methods, such as Logistic Regression, struggle to capture the complex semantic and contextual dependencies in informal and diverse text, often leading to suboptimal performance. To address these limitations, we propose a novel approach that combines BERT with BiLSTM networks and utilizes data augmentation techniques. BERTs pre-trained embeddings contain rich contextual information, while the BiLSTM layer models long-range dependencies within the text. Data augmentation is applied to expand the training dataset, improving the model's generalization and robustness. Experimental results on the ACL IMDB dataset demonstrate that our approach outperforms traditional models, with the BERT + BiLSTM model achieving a testing accuracy of 88%, compared to 82% for BERT alone. This study shows that combining BERT, BiLSTM, and data augmentation significantly enhances the accuracy and robustness of emotion classification models, particularly for complex and informal text.