Research On Stock Index Forecasting Driven By Multi-document And Historical Transaction Data

The stock market can reflect the development of the social economy and the company,and it is called the "barometer" of the social economy.In today's economic globalization,if the stock market can be predicted in advance,it is of great significance for the country's macroeconomic regulation and control.At the same time,it helps to improve the government's ability to supervise the financial market,thereby maintaining the stable development of the financial market.However,the forecast of the stock market depends on many factors,and the stock market is uncertain and volatile,making it a challenging task to explore and master the laws of the stock market.The stock index can fully reflect the overall trend of the stock market.Therefore,analyzing the stock index and making post-mortem predictions is of great significance for investors,companies,and the country to make decisions.It has also become a major player in the financial and computer fields.A main research content.The influencing factors of the stock market can be attributed to two aspects.On the one hand,it is affected by the policies and regulations issued by the state,formal news documents,and social media posts and comment documents(hereinafter referred to as multi-documents)issued by investors.This is affected by historical trading data of stocks.This topic proposes to integrate the above two information sources(multi-document and stock historical transaction data)to predict the rise,fall and closing price of the Shanghai Stock Exchange Index.According to the characteristics of documents from different sources,this paper proposes shallow feature extraction methods for different documents,and then uses autoencoders to perform deep feature extraction on shallow feature vectors in order to dig out hidden abstract features in the text.Finally,the characteristics of text and stock historical data are combined to realize the prediction of stock index.This article explores the effectiveness of multi-document deep feature extraction,studies the impact of "time step" on stock index prediction,and compares the accuracy of stock index prediction under the following two inputs:(1)Only multiple documents are used,(2)Combine multiple documents and stock historical trading data.The experimental results show that: the deep feature extraction of multiple documents is easier to mine the hidden features in the documents;the stock market is delayed;the combination of multiple documents and historical stock transaction data can indeed improve the prediction accuracy of stock indexes,and the model constructed in this paper is in Both the stock index rise and fall forecasts and the closing price forecasts can achieve better accuracy,which has certain guiding significance for investment and financial management.