BERT-Based Depressive Sentiment Analysis and Explainability for Chinese Texts

The extensive application of NLP technologies has greatly increased in recent years, for instance, in translation, text classification, and text generation. With the advancement of technology, models have become more complex, which raises the difficulty of model explanation. The paper explores sentiment analysis classification for identifying depressive text using the BERT model and demonstrates the explanation of model outputs through explainable artificial intelligence (XAI) technology. The pre-trained BERT model was fine-tuned by using a Chinese text dataset, which consists of Chinese texts with depression labels obtained from social media like Weibo, REDnote, and Twitter. The paper reveals that the deep learning model, leveraging contextual information and interpretability techniques, effectively analyzes depressive sentiment, and has endowed the model with certain interpretability.