A Research And Design Of Stock Market Risk Prediction Model Based On Sentiment Analysis

With the rising of living standard,more and more people are willing to participate in investment activities,of which the stock market has become the first choice for ordinary investors because of its low entry threshold and considerable returns.Behavioral economics holds that people's irrational behavior is an important factor affecting the stock market.With the popularity of the Internet,many investors like to express their views and opinions through the network.Through in-depth mining of these data,we can understand the emotional tendency of market participants,so as to provide help for the stock market decision-making,and provide reference for investors to reduce the risk.Based on this idea,this thesis launches the following research.Firstly,this thesis introduces a method to extract text feature words based on word popularity to handle the news commentary of specialized field.This method fully excavates the timeliness of news text,so as to make up for the limitations of traditional methods in short text.The advantages and feasibility of the method are proved by experiments.Secondly,this thesis proposes a clustering system to extend the traditional sentiment lexicon in view of the lack of sentiment lexicon of specialized field and external labeled data.Then,the extended sentiment lexicon is used to complete the text sentiment classification based on machine learning method.Finally,according to the news commentary sentiment we calculate daily sentiment index,and analyze the relation between the performance of the stock market and the sentiment index based on the real market data.Then,a stock market risk prediction model based on supervised learning method is designed,and the validity of the model is verified by experiments.The results show that the sentiment index obtained from the news commentary is closely related with the parameters of the stock market,and it is feasible to forecast the stock market risk through the sentiment index.At the same time,this thesis presents a practical method for sentiment analysis in the absence of external labeled data and a sentiment lexicon for specialized field.