A Research On The Sentiment Analysis Methods Of Stock Comments

With the popularity of the Internet,the increase of Internet users,the community news will be spreaded in the first time of the Internet.The speed of stock information is faster and more influential.Many people always prefer to look up stock news which they concern before making the investment decision.But it is not easy to find real valuable messages among the massive information,so it is necessary to get an emotion analysis to quickly get the emotional tendencies of the stock comments,which will help investors,experts,companies and other great convenience.In this paper,we analyzed the characteristics of the stock comments,and compared different methods to find a suitable emotion analysis algorithm,which help people quickly and accurately access useful information.Based on emotion analysis and emotional information extraction technology,this paper studies the sentence-level emotion analysis method of stock comments.Through the analysis of the text of the stock review,it can be divided into two kinds of situation which are single stock and multiple stocks.The single stock can be directly analyzed with the sentence,but the latter one needs to identify the name of each stock and its corresponding evaluation of words first,and then emotion analysis could be started.According to the comments of every single stock,the paper studies the short and long term of the stock respectively.The data needs to be processed first before analyzing such as cleaning,labeling and so on.Secondly,we compared the results under different dimensions and feature weight selection methods by applying methods like SVM,KNN,Bayesian and other machine learning algorithms.It shows that with feature dimension of 300,SVM is the best performance in these methods.To those coments which includes multiple stocks,pre-processing the data is also the first step of studying.After the data gets prepared,we use methods based on keyword extraction and syntax of the rules to extract the comment unit,which includes the name of the stock and its evaluation words,and apply SVM and methods based on emotional dictionary to classify the comment units.As the conclusion,we found that rules based on the dependent syntax can identify the evaluation unit in the stock index.Besides,generally speaking,the classification effect of the emotional analysis on the evaluation unit with the emotional dictionary is good.