search for




 

Predicting Korean stock prices using heterogeneous dynamic graph neural networks (HDGNN)
Journal of the Korean Data & Information Science Society 2025;36:13-22
Published online January 31, 2025;  https://doi.org/10.7465/jkdi.2025.36.1.13
© 2025 Korean Data and Information Science Society.

Minyoung Yoon1 · Yongku KIm2

1Data Science, Kyungpook National University
2Department of Statistics, Kyungpook National University
Correspondence to: This research was conducted with the support of the National Research Foundation of Korea(NRF) funded by the Ministry of Education (RS-2023-00240494) and Ministry of Science and ICT(RS-2023-00242528).
1 Graduate student, Department of Data Science, Kyungpook National University, Daegu 41566, Korea.
2 Corresponding author: Professor, Department of Statistics, Kyungpook National University, Daegu 41566, Korea. E-mail: kim.1252@knu.ac.kr
Received November 4, 2024; Revised November 26, 2024; Accepted November 27, 2024.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
This study designed a stock price prediction model for the Korean stock market using a heterogeneous dynamic graph neural network (HDGNN). To effectively capture the nonlinear characteristics of stock prices and various influencing factors, we utilized a dynamic graph structure that integrates data from stocks, economic indicators, news, and disclosures. Using daily data from January 2022 to December 2023 for KOSPI 200 stocks, we established relationships between nodes based on correlations and performed topic extraction with LDA and text embedding with FinBERT for news and disclosures. Temporal embeddings were generated using LSTM to reflect each node’s time-series characteristics, while multi-level attention mechanisms were applied to integrate information at the node, temporal, and graph levels. Experimental results demonstrated that the HDGNN model achieved superior performance in MAE and RMSE metrics compared to existing models such as RNN, LSTM, and GRU, thereby enhancing the accuracy of stock price prediction.
Keywords : Attention mechanism, heterogeneous dynamic graph neural network, stock price prediction, time series embedding