Construction And Application Of Investor Sentiment Index Based On RoBERTa Model

Since the formation of the securities market,the changing trend of stock prices has always been the most concerned content of securities investors.Behavioral finance has the concept of investor sentiment,which can explain various financial anomalies that cannot be explained by traditional finance,and plays an important role in evaluating the risks and returns of the stock market.At present,with the rapid development of the Internet and big data technology,there is a large amount of investor comment data in all corners of the Internet,which has become the best data source for the construction of investor sentiment index.Using natural language processing technology to classify the sentiment of investor comments,and constructing the investor sentiment index based on the classification results,has become the most mainstream research method of investor sentiment.The existing investor sentiment index construction method mainly uses the number of positive and negative comments in investor comments,and does not take into account the impact of neutral comments.Lowering and the tendency of investor comments to be colloquial and casual,the influence of neutral comments on investor sentiment can no longer be ignored.Therefore,this paper proposes a new formula for investor sentiment index considering neutral comments.The investor sentiment index considering the number of neutral comments has higher requirements on the accuracy of sentiment classification.In this paper,the Ro BERTa model,which is the most advanced in the field of natural language processing,is selected as the sentiment classifier,and the crawler program is used to obtain the comment data of the Dongfang Fortune Internet Bar.Investor sentiment index is constructed by training the model and using investor comments of individual stocks over a period of time.After comparison,the Ro BERTa model has a better classification effect than the original BERT model and the Fin BERT model based on financial texts.Then,through Granger causality analysis and impulse response graphs,this paper confirms that there is a two-way Granger causality between investor sentiment index and stock return,and there is a clear mutual influence within about 7trades.Finally,this paper combines investor sentiment with stock technical indicators and indirect indicators that reflect investor sentiment,and uses LSTM neural network to build a stock rise and fall prediction model,which achieves an accuracy of 67.06%.We also use the trading stategy to vertify the application possibility of the prediction model.The research results obtained in this paper can not be used as a standard for investors to buy and sell stocks,but they can still provide suggestions for investors to make investment decisions.The government,securities industry associations and other institutions can also make timely stability maintenance policies based on the relationship between investor sentiment and stock price fluctuations to prevent stock market disasters and ensure the stable development of my country’s securities market.