Research On User Portrait Generation And Behavior Prediction For Securities Investment

With the gradual maturity of China's financial market,the proportion of individual investors has reached 99.76%.A large number of individual investors need the guidance and protection of the state,and also need financial institutions to provide personalized financial services.Therefore,it has become an important research topic to comprehensively model and extract various features of investment users and effectively predict their investment behavior.Traditional investment user modeling and analysis methods rely on questionnaires,lack of objectivity,and do not consider the key impact of real-time changing market environment on user investment behavior.In view of the above problems,this paper proposes a securities investment user analysis modeling framework,which can generate comprehensive and interpretable user profile feature description for investment users,and effectively predict users' dynamic investment behavior.In view of the lack of user data in the field of securities investment,this paper designs a web crawler based on scrapy framework to obtain basic user data,simulated transaction data and stock bar posting data in financial service websites.In order to analyze the market environment faced by users in the process of investment,we also obtain the stock market and attribute data at the same time,and ensure the data integrity and accuracy through preprocessing,and finally form the securities investment user knowledge base.Starting from the user's investment behavior,this paper combines the user's investment data with the market data,and proposes a set of security investment user profile label system and generation method.Four types of user tags are established,among which the investment style labels are analyzed from the aspects of user's operation preference,trading opportunity and variety preference.Aiming at the emotional information tags,this paper analyzes the sentiment of the text information posted by the user's stock bar based on the LSTM model.The effectiveness of user profile in describing the characteristics of investment users is proved by user clustering experiment.At the same time,based on the user's needs for understanding themselves,the user profile is implemented including text description and chart display.Based on the idea of representation learning,this paper proposes a decomposition method of stock trend perception matrix,which can obtain the intrinsic attributes of stocks and the potential investment preferences of users.A deep stock attention network based on user profile(DSANP)model is constructed to predict user investment behavior.Attention mechanism is introduced to effectively integrate the dynamic index sequence representation of stocks with user profile features to obtain users' dynamic perception of stock historical quotations.Compared with other users' investment behavior prediction model,DSANP has achieved more than 6%improvement in the comprehensive evaluation index Macro-F1,which verifies the effectiveness of the proposed investment behavior prediction model.