Research On User Investment Behavior Modeling And Attribution Analysis Method

After decades of development,China's securities market has been continuously improved,with a large number of individual investors participating in securities investment,and a large number of investment behavior data have been accumulated in the investment process.However,most investors do not have professional investment knowledge and cannot make accurate analysis and evaluation of their investment behavior.However,an objective and scientific evaluation of investors' investment behavior can help investors improve their investment ability.At present,the research object of securities investment evaluation is mainly funds.Due to the great difference between individual investment behavior and funds,some existing performance attribution models cannot be applied to large-user investment behavior analysis.User modeling is one of the important directions of user behavior analysis.This topic takes user investment behavior as the research object,proposes a user performance attribution model,and studies the application of deep neural network model to user investment behavior modeling analysis.The research work of this paper includes the following research contents:For the traditional machine learning algorithm,only the statistical characteristics of the user's investment behavior are used,and the other rich information contained in the user's investment behavior cannot be processed,such as time dynamic information and user behavior preference.In this paper,RNN network,which is widely used in processing serialized data,is introduced into the modeling of user's investment behavior,and the model is trained in the task of investment behavior prediction.The network is analyzed and optimized according to the user's investment behavior data set.The introduction of attention mechanism into the common behavior modeling framework can make the model pay attention to more representative user investment behavior in the process of modeling user investment behavior.Since the above model only uses the sequence information of user's investment behavior in the process of modeling user's investment behavior,only the user's behavior pattern or user preference information after time series dynamic data is learned,but the importance of relevant user information is ignored.In this paper,we further study the introduction of new gating units for LSTM neural units to introduce user information.After the introduction of user information,the performance of the model on the evaluation indicators has been improved.The introduction of the attention mechanism may make the neural network model pay too much attention to some behaviors in the user's investment behavior sequence.The model pays too much attention to the local information and ignores the global information of the sequence,so that the information learned by the model is too biased to fully reflect the user behavior recording.In this paper,the model framework of parallel joint training is used to solve the problem.The performances of the models based on the original LSTM neural unit and the improved LSTM neural unit are improved.In order to solve the problem that the existing Brinson performance attribution analysis model cannot be applied to the analysis of investment behavior of individual investors,this paper redefines the calculation method of the actual return on user investment,and on this basis,proposes a user performance attribution model suitable for individual investors.Based on user performance attribution model,a system for quantitative analysis and evaluation of user investment behavior is built.