Internet Financial Transaction Algorithm Based on Graph Deep Learning

This paper focuses on exploring the application of Graph Neural Network (GNN) algorithm based on graph deep learning in Internet financial transaction algorithms. In the field of Internet finance, traditional trading algorithms are faced with problems such as limited data processing ability, neglect of market structure, and insufficient generalization ability. The purpose of this study is to improve the performance of financial trading algorithms in processing complex data, understanding market structure and improving generalization ability through graph deep learning technology. This paper begins with a review of the limitations of traditional financial transaction algorithms, and then summarizes the main research advances in this field and analyzes the shortcomings of existing research. This paper introduces in detail the graph deep learning method adopted in this research, namely GNN algorithm, and how to improve the effect of trading strategy through this method, and designs three experiments for this purpose. In the historical data backtest experiment, the graph-based deep learning GNN algorithm achieved a higher annual return rate than the traditional algorithm, reaching 12%, and the maximum retracement index data was 8%. In the real-time market experiment, the cumulative return of GNN algorithm is 6% and the standard deviation of volatility is 1.2%. In a third anti-stress experiment, the GNN algorithm showed greater stability and risk control under extreme market conditions. The experimental data results show that GNN algorithm has significant advantages compared with traditional algorithms in Internet financial transactions.