A Method to Classify Emotions Based on BERT

Nowadays, as science and technology continue to advance, addressing people's emotional issues is of particular significance, and how to accurately and efficiently recognize emotions has become increasingly prominent. We explores sentiment recognition and classification methods, including rule-based approaches, traditional machine learning models, deep learning models, and multi-modal sentiment analysis. Through comparative analysis, we will evaluate the effectiveness of these methods in handling social media texts. Additionally, we has developed a BERT-based two-stage model that first determines sentiment polarity (positive or negative) and then performs detailed classification of negative emotions. This two-stage model significantly improves accuracy and detail recognition compared to traditional one-stage models and other baseline methods. The results show that our approach not only enhances sentiment analysis accuracy but also provides deeper insights into the emotional subtleties of social media texts. This advancement holds significant practical implications for real-time sentiment monitoring and more effective crisis management strategies.