ST-HGSL: A Spatio-Temporal Hypergraph Structure Learning Framework for Equity Factor Prediction

View at publisher Summarise with VIRAT
Authors: Junling Zhu, Tianren Chen, Qing Tian, Yihe Tan, Yujie Lin
Cited by: 0
References: 12

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Predicting stability in complex networked systems (such as financial markets) is a critical yet challenging task for AI system security. Existing methods predominantly rely on sequential models that treat nodes in isolation or standard Graph Neural Networks (GNNs) based on static priors. While recent hypergraph approaches attempt to capture high-order correlations, they often depend on pre-defined generation rules that fail to adapt to rapid system regime shifts. To address these limitations, we propose a novel framework named ST-HGSL (Spatio-Temporal Hypergraph Structure Learning). Specifically, we employ a Causal Temporal Convolutional Network (TCN) to extract high-frequency temporal features while strictly preventing future information leakage. Simultaneously, a differentiable Hypergraph Structure Learning module is designed to automatically discover implicit structural dependencies purely from data. A Hypergraph Neural Network (HGNN) is then utilized to model the spatial propagation of risk within these learned high-order structures. Extensive experiments on the CSI100 dataset (2022–2025) demonstrate that ST-HGSL achieves state-of-the-art performance with an Information Coefficient (IC) exceeding 0.031. These results validate the model’s significance as a robust AI framework for detecting system-wide instability in dynamic networked environments.

Fields of Study & Subjects

Stock Market Forecasting Methods Advanced Graph Neural Networks Complex Network Analysis Techniques

Publication details

Journal / source
International Conference on Artificial Intelligence and Social Network Systems (AISNS)
Pages
79–83
Publication date
9 January 2026
Language
EN