Research On Stock Trend Prediction Based On Text Sentiment Analysis And Improved SVM

With the development of online stock trading and information exchange and sharing,more and more stock investors express their views on the stock in the stock forum according to the judgment and prediction of the state's public policies and the company's future development prospects,which has become an important factor affecting the rise and fall of the stock.This paper takes the real-time stock post of the stock bar of East Fortune online stock exchange as the research object,and analyzes the text information by using natural language processing technology.It reduces the feature dimension by feature engineering technology,optimizes the SVM model by grid search,and establishes the stock price rise and fall trend model of Shanghai stock index.The research contents of this paper include text mining,financial sentiment dictionary construction,sentiment index calculation and SVM model improvement based on feature engineering technology and grid search method.First of all,this paper crawls the text comments in the East Fortune Shanghai stock index bar as the text corpus.Secondly,the traditional basic sentiment dictionary is expanded to increase the stock market vocabulary and network sentiment vocabulary in the common words in the financial field,so as to improve the accuracy of sentiment analysis.Then,the natural language processing technology is used to preprocess the text comments,and the sentiment index of the comments is calculated by matching the semantic rules.Through the verification,we find that there is a positive correlation between the sentiment index and the stock price.Then,PCA and GA algorithms are used to extract and select the features of the input feature variables,which simplifies the complexity of the model and reduces the amount of calculation.Finally,the parameters of SVM model are optimized according to grid search and4-fold cross validation.Through comparative analysis of the classification and prediction performance of six models,the effectiveness of the SVM model based on text sentiment analysis,GA feature selection and grid search(Sentiment Index+GA-GS-SVM)in the stock index rise and fall prediction problem is fully verified.The final model Accuracy = 59.01%,Precision = 54.15%,Recall = 62.16%,F1 = 57.88%,AUC = 0.6042.It shows significant classification performance advantages,and can provide certain reference and help for the majority of investors.