| As people’s lives become deeply integrated with the Internet,the stock market has attracted many individual investors,the vast majority of whom trade stocks using the Internet.Posting stock comments on stock forums has become a popular phenomenon,which generates a large amount of stock commentary.The sentiment of investors contained in this information can affect the trend of the stock market.In recent years,with the development of natural language processing technologies,research on text sentiment analysis has been continuously deepening.However,due to the lack of large amounts of labeled data,traditional domestic stock sentiment classification methods are difficult to improve accuracy,which poses a challenge for stock market researchers to obtain investor sentiment.This paper fully considers the limited labeled data of stock comments and proposes a BERT-based domestic stock sentiment analysis model.The specific research content is as follows:(1)To solve the problem that the limited labeled data of stock comments makes it impossible to improve the accuracy of sentiment classification,this paper considers introducing a pre-training model based on BERT,which can effectively obtain semantic expression of text on a large amount of unlabeled data,and designs a new model structure.(2)To obtain the most optimal model,this paper constructs a complete experimental plan,proposes three different variants based on BERT as the experimental group to obtain relatively excellent model results,and designs and selects five characteristic models as the control group to obtain the optimal model by comparison.The BERT+FC model proposed in this paper has better performance than traditional machine learning methods and neural network methods.Empirical results show that the stock comment sentiment analysis result is effective,and the model has strong generalization ability to perform sentiment analysis in multiple fields. |
