Research And Application Of Financial Text Sentiment Analysis Method Based On Deep Learning

With the increasing maturity of Internet finance and securities markets,investors are constantly pouring into the stock trading system to generate massive amounts of data all the time.Artificial intelligence and deep learning technologies have been widely used in the financial industry.According to the non-efficient theoretical model of the market,that is,stock prices do not fully reflect the real value of listed companies,there is a certain correlation between market sentiment and stock price trends.Therefore,sentiment analysis of financial texts can be used to help investors make judgments that are in line with the market.This paper takes the investor sentiment information in the stock bar as an example,and combines deep learning technology to conduct the following research and exploration.(1)Aiming at the defect that the recurrent neural network analysis model ignores sentiment words,this paper proposes a sentiment analysis method based on Fin BERTBi ALSTM.This method uses Fin BERT to obtain dynamic word vectors of text and eliminates the effect of polysemy,and transfers the obtained word vectors to Bi LSTM model to enhance text sequence information.By applying the attention mechanism to enhance the weight of feature words and integrate text semantics,the ability to learn long-sequence word order is enhanced,and the model’s discrimination accuracy for financial texts is effectively improved.(2)Aiming at the problem that convolutional neural networks tend to ignore contextual semantic information,this paper proposes a multi-channel text sentiment analysis method based on NT-FABi LSTM.The Fin BERT model is used to extract text word vectors,and the improved Text CNN and Bi ALSTM models are used to extract the local information and global information of vector features respectively,and the two vectors are fused through the fusion layer to achieve sentiment classification in the fully connected layer.The experimental results show that the accuracy rate and F1 value and other indicators have increased.(3)Based on the above sentiment analysis model,a stock trend prediction model integrating text sentiment analysis and stock historical market data is constructed.Taking sentiment classification as the parameter,an indicator system for investor sentiment expression is designed,which integrates investor sentiment and historical market characteristics.The experimental results show the feasibility and practicability of the combination model proposed in this paper in the task of predicting domestic stock price trends.