Research And Implementation Of Stock Index Prediction Technology Based On Sentiment Analysis

The investor sentiment hypothesis in behavioral finance tells us that emotions can have an impact on their decision-making,and in the relative experiments it verifies that the investor's emotions will have an effect on the market.With the arrival of big data era,social network has become a popular venue for sharing opinions and emotional expression,and making a comprehensive and effective quantitative analysis of massive comments can guide for stock index prediction.So making a research about the technology of stock index prediction based on sentiment analysis will be important for exploring the law of stock market volatility.Considering the traditional numerical prediction model can't take the mood volatility caused by news into consideration,this article needs construct a prediction model that can merge text comments with historical numerical data to do stock market analysis.This article's blog comments is mainly collected from Sina microblog,and we put forward a sentiment classification algorithm of fast construction of stock-oriented sentiment lexicons,then compare it with algorithm based on supervised learning,the result shows that the lexicons method achieved the best consistent accuracy,up to 94%,and the second is logistical regression(LR)algorithm,which accuracy is 90%.Based on the already determined sentiment classification algorithm,this article makes a correlation analysis about online comments and Shanghai Composite Index(SCI),it mainly uses Pearson method to do visualization and correlation analysis between different sentiment index and SCI closing value.The research result shows that the max coefficient is up to 0.91 and the average value is 0.55,which indicates that investor sentiment has some correlation with SCI closing value.Based on the sentiment classification research and correlation analysis,this article puts forward an algorithm which merges sentiment index with regression model to do SCI closing value prediction.Specifically,this article merges four kinds of sentiment index which is computed by two best classifiers with SCI 5-min closing values to construct multi-linear regression prediction model,and compares it with one not using sentiment index.The result shows that the first type achieved higher accuracy,and among the first type the one using sentiment index computed by LR achieved best.At the end of this article,we not only make a description about the limitations of above research,but also talk about the challenges when putting sentiment analysis into actual investment.