Stock Prediction Research Combining Sentiment Analysis

Stock investment is a very active way of investing and managing money.Investors' trading behaviors in stock market are all intended to make a profit.At present,stock predictions mostly are based on the historical data of stock trading.This paper studies stock prediction models combined with the sentiment analysis of stock comments.These models analyze the data of sentiment tendency,stock trading index,time series and so on.Sentiment analysis:The comment text data of particular stock used for analyzing are come from active stock forums.These forum data are a large number of short texts including noise,reflecting the views of small and medium investors.Using the SVM classifier,based on the LIBSVM toolkit of JAVA version for text classification,to obtain the emotional tendency index Bs through calculating and analyzing.Improvement:Analyzing the data of different periods to establish BP network prediction models,and calculating the different influence values of the data in different periods according to the MIV algorithm.While calculating text emotion tendency index Bs,combining the impact weight of the text authors.Model design:The BP neural network model has only five stock trading indicators used as inputs,as the reference model,marked as model-1;The BP network model combine sentiment index with five stock indicators used as inputs,denoted as model-2;The BP network prediction model analyze the closing price of the five trading days before forecast day,denoted as model-3.Conclusions:The prediction accuracy of model-2 which include sentiment index is higher than the accuracy of model-1;The model-3 combining the MIV algorithm,obtain different impact values of the five trading days before prediction day,the result conforms to the tendency that the data closer to prediction day have larger weight of influence.