Research On Deep Learning Based Event-driven Stock Prediction

The global economy is at a stage of rapid development,and China's financial market is also constantly developing.High returns in financial markets attract various types of investors.Financial markets have dynamic continuity and are changing all the time.Behind high returns is high risk.Investors are always paying attention to the dynamics of financial markets.Media reports on the market will affect investors' investment decisions,and may even cause market sentiment fluctuations,and the price movement of the asset.There are a large number of news events reported every day.If one can identify the target of affected financial assets from a large number of news events,and predict the price of the assets,it is meaningful to investors.In previous studies on the stock market,researchers often overlooked news events that were not targeted by specific entities.Many studies based on known stocks under the premise of predicting stock prices.This article defines news events that are not targeted by specific stock entities as “general events”.For example,news events reported in the real estate industry may not mention related stock entities,but experienced investors can pick out the stocks that may be affected in a very short period of time.This paper starts with the attribute relationship between events and stock entities,and proposes a new deep learning model(MEAN)that can quantify the correlation between events and stock entities.Through the multi-path association calculation,the MEAN model fully exploits the correlation between stock attributes and news events from multiple dimensions,and finally outputs the correlation coefficients of the two through the fully connected network.The second innovation of this paper is to propose a heterogeneous information collaboration model(HICN),which combines stock price information and news event information to predict the stock price of the stock set anchored by the MEAN model.MEAN and HICN constitute the event stock price prediction model of this topic.This paper designs several control experiments based on experimental data.The control model includes traditional machine learning model and commonly used deep learning model.The validity of the event-driven stock price forecasting model based on deep learning is verified by the comparison of experimental results.In this paper,the simulation investment strategy is constructed according to the prediction of the model,and the excess return of 27% of the relative market is obtained,which indicates that the model has guiding significance for actual investment.