Research On The Application Of Multi-source Data Fusion And Deep Learning Technology In Stock Market

As one of the important components of the financial market,the stock market is often regarded as a barometer of the economy because it reflects the trend of economic changes.The forecast of the stock market has important reference value for investors' rational investment and government macro-control.However,due to uncertain factors leading to dynamic changes in the market,stock forecasts have become very challenging.Deep learning is an important research branch in the field of machine learning.Compared with traditional statistics and machine learning models,deep learning can handle the multivariate stock market because of its good processing ability on nonlinear data and mixed information.The non-stationary and non-linear dynamic data can get more reliable analysis and prediction results.With the development of information technology,data types are becoming more and more abundant,which provides conditions for stock prediction results to become more accurate,but at the same time it is more difficult to include these data for processing.The feature engineering of this article will construct data features from four aspects: stock trading data,time-frequency features based on trading data,news text data and expert factors.(1)The trend of stocks is inherently related to the historical trading data of stocks.When constructing data features,the trading data of stocks is taken as part of the features.(2)Considering that historical stock transaction data is a non-stationary sequence,it is difficult to directly use non-stationary data to make predictions.In feature engineering,non-stationary stock transaction data is transformed into a stationary sequence by means of non-stationary sequence decomposition.feature.(3)In addition,stocks will be affected by fundamental conditions(policies,company operating conditions,etc.).Data features will also introduce news text information,use sentiment analysis tools to extract sentiment factors in news texts,and add sentiment factors to data features in.(4)Data features include transaction data features,time-frequency features,and news sentiment features.Considering that the industry will use some expert factors when predicting stocks,the feature engineering introduces two expert factors,Alpha101 and Alpha191.Data characteristics.Finally,the characteristics of the above four aspects are standardized and fused to construct a multi-source heterogeneous feature.In this paper,while constructing multi-source heterogeneous data features to improve input features,by improving the Generative Adversarial Networks,it solves the problem that statistical methods cannot handle non-stationary and non-linear data well,and the parameter updates of traditional deep learning methods come from data samples rather than the relationship between the data.Therefore,the improved Generative Adversarial Network can better explore the law between stock data.Finally,comparing the traditional method and the experimental analysis of the improved Generative Adversarial Network,it is found that the improved stock prediction model based on the Generative Adversarial Network has a better effect in the field of financial stocks.