StockGraphPredict: a temporal graph convolutional network with attention mechanism for stock prediction

Abstract

Stock trend prediction has been a major focus in both academia and industry. The advancement of computer technology has led to the use of neural network techniques like LSTM, RNN, and SVM for predicting stock prices based on historical data.

However, stock price fluctuations are influenced by various factors, and relying on a single data source often fails to capture the full market dynamics. The relationships between companies, particularly within the same industry, significantly impact stock prices. As such, integrating multiple data sources and deep learning methods has become a key development in stock prediction.

Moreover, the stock market has complex spatiotemporal dependencies, where stock relationships are influenced by industry, region, and economic factors. Combining Graph Neural Networks (GNN) and temporal deep learning models can effectively uncover correlations between stocks and improve prediction accuracy. This study aims to enhance stock trend prediction by using a fusion model based on deep learning, considering both historical data and stock interrelations.