Research On Security Transaction Risk Identification Based On Graph Convolutional Neural Network

Securities investment trading is a trading behavior in which securities analysts who have a deep understanding of the stock market conduct stock trading operations after predicting the future development direction of the stock market and the degree of rise and fall based on the development of the stock market.Among them,stock prices are time-series dynamic,random,and high-noise.In addition to the massive data generated by the financial market in the mobile Internet era,traditional financial transaction analysis methods are obviously insufficient.For example,in terms of stock forecasting,compared with graph neural networks and time-series neural networks The network method has great limitations.Deep learning algorithm is an important method to dynamically mine stock price trend information from massive,multi-dimensional,multi-dimensional,and heterogeneous financial data.Traditional methods realize dynamic monitoring and forecasting of stock prices through modeling and analysis of data on stock market influencing factors.There are four main shortcomings: First,there are a lot of data that affect securities trading and stock market fluctuations,the particles are relatively coarse,and the quality is not high.The collection burden is heavy;second,there are many indicators for the risk evaluation of securities trading and stock market volatility;third,the existing deep learning algorithms are prone to overfitting due to the failure to fit the data characteristics in the securities trading and stock forecasting And the phenomenon of underfitting.Fourth,the existing algorithms fail to make full use of the timing characteristics of the data.The main research work of this paper is as follows:1.Aiming at numerous risk evaluation indicators for securities trading,we use the random forest algorithm to optimize the noise reduction of the primary indicators.Then,the algorithm automatically analyzes the primary selection indicators and iteratively calculates the importance of features,and selects the indicators that have a more significant impact on the securities transaction risk as to the securities transaction risk identification indicators.2.Aiming at the problem of failing to make full use of the time series characteristics of the data,we introduce the long-term dependence relationship between the time step sequence data of the time series memory network learning.Then,it constructs a complex and deeper function between the risk identification and evaluation index and the risk of securities trading.3.Aiming at failing to fit the data characteristics better,we propose a neural network of time series theme graph(Neural network of time series theme graph,referred to as TSTNet)with adjacency matrix independence test.The TSTNet algorithm model obtains the three-level index characteristics of the stock through the time series theme,and then is fused with the correlation coefficient matrix parameters through the principal component analysis(PCA)picker.Finally,according to the natural language word frequency embedded analysis feature value,the coefficient matrix between the independent variables is calculated through the model to obtain the early warning interval value.This paper uses the stock market data set(2010-2020)as the test sample.The experimental results show that the accuracy of the proposed TSTNet model is 0.8589,the recall rate is 0.8591,the coverage rate is 0.8902,and the MAE is 0.6303,which indicates that the model has a high The accuracy of risk identification can more accurately recommend investable securities with high economic returns to investors,effectively reduce the probability of securities investment risks,and effectively increase investors’ investment returns.