Practical Applications of ChatGPT and BERT: Long Text Summarization and Model Feature Ablation Study for Stock Comment Topic Identification

Topic identification of expert stock comments is an important task in the field of finance, which helps investors understand the market dynamics and expert opinions so that they can make more informed investment decisions. The Bidirectional Encoder Representations from Transformers (BERT) model is one of the widely used methods for text topic identification. To improve the identification accuracy and overcome the limitations of the BERT model, two strategies are proposed in this study. First, given the BERT model's input length limitation when dealing with long texts, we used the summaries created by Chat Generative Pre-training Transformer (ChatGPT) as inputs to the model. The generative summarization capability of ChatGPT not only reduces the length of the text effectively but also preserves the key information of the original text, which enables the model to better understand and deal with long texts. Second, we perform multi-layer feature ablation on the BERT model to find the most representative features for text topic identification. The multi-layer features of the BERT model can be more effectively utilized to improve the accuracy of text topic identification. Experiments show that our proposed strategies exhibit excellent performance and provide effective paths to improve the accuracy of stock comment topic identification.