BERT-RCNN-SE: A Fine-Grained Sentiment Analysis Model for Social Media Text Comments

Sentiment analysis plays a crucial role in tracking opinions on social media and gaining insights into user sentiment. However, traditional methods mainly focus on the overall sentiment of the text and struggle to identify nuanced emotions. This study presents a fine-grained sentiment analysis model called BERT-RCNN-SE, designed to address the immediacy, irregularity, and complexity of sentiment expression on social media platforms, such as Twitter. The model utilized BERT for semantic representation and the RCNN architecture to extract detailed local aspects of phrases. Furthermore, the SENet module adaptsively modifies the weights of feature channels, improving the model's ability to recognize nuanced emotions in social media comments. The results indicate that the BERT-RCNN-SE model has significantly outperformed the TextCNN model on a publicly accessible Twitter dataset, with a Macro F1 score improvement of 38.02%, compared to the baseline BERT model which improved by 1.15%. This confirms the model's effectiveness and superiority in fine-grained sentiment analysis of social media comments.