Stock Trend Forecasting And Recommendation System Based On Multiple Types Of Features

With the development of artificial intelligence and economy,more and more smart stock investment products have appeared on the market.These products are based on artificial intelligence algorithms that provide users with stock recommendation and other functions.If we can design effective stock forecasting and stock recommendation algorithms,it will further enhance the actual user experience of smart stock investment products,and will also bring progress to both artificial intelligence and economic development.To this end,the following work has been carried out in this paper.Considering that the rise and fall of A-shares is very susceptible to many factors,this paper combines several types of features,including stock company's fundamental data and financial data,as well as stock market data.Due to the large number of features,it is necessary to make reasonable and effective feature selection for various features to reduce the dimension of the data,extract important features and discard features with less significance and introduce noise.In this paper,we will experiment with a variety of feature selection methods to find out the most effective feature selection method for stock forecasting.In addition,in order to model the stock data,this paper designs a series of feature engineering work for stock data sets,and constructs a data set suitable for modeling.In terms of models,combining multiple deep learning methods can overcome many of the difficulties and limitations faced by traditional forecasting methods while avoiding the influence of human factors.This paper first uses traditional machine learning methods(SVM,LR,and DNN)to model the non-timing features of stocks,and uses RNN and CNN model to model the timing features of stocks.Then,this paper proposes a deep learning model combining timing and non-timing model parts.The model can handle non-timing features and timing features simultaneously,with the "Wide" portion responsible for handling non-timing features and the "Deep" portion for processing timing features.The model then models the features extracted from the two parts.The model is superior to other basic models in evaluation criteria such as f1 values.In order to carry out stock recommendation more accurately,based on the abovemodel,this paper constructs a classification data set and a regression data set respectively.After modeling the two data sets respectively,using the model ensemble method,the results of the two base models are continuously modeled by the strong classifier random forest,and the obtained base classifier weights are used to calculate the recommended scores of the stocks.Based on the recommendation score,the stocks are ranked and recommended.On the basis of the traditional evaluation of the recommendation,the paper also calculates the investment evaluation indicators such as the rate of return and the excess return rate through the simulation transaction,which makes the recommendation result more realistic.In addition,in order to allow users to operate the entire recommendation and simulation transaction process,this article builds a stock recommendation system based on the Django framework.The user can define the recommended time period and recommended stock number that he likes,and then the system will automatically recommend the corresponding stock for it and conduct a simulated transaction.The recommendation system will display the results of the stock recommendation.