Dynamic-aware hypergraph learning for stock recommendation

Abstract

The stock market is influenced by multiple factors, making stock recommendation a challenging task. In recent years, several studies have improved recommendation performance by modeling higher-order relationships between stocks and industries through hypergraphs. However, the existing models still face numerous challenges: such as relying solely on a single stock feature, failing to account for the dynamic influence of the industry on individual stocks, and overlooking the heterogeneity in stock performance within the same industry. To address these challenges, a Dynamic-Aware Hypergraph Learning (DAHL) is proposed for stock recommendation. Firstly, the model designs a dynamic-aware module that captures the diverse and dynamic relationships between industries and individual stocks through an attention mechanism, emphasizing the heterogeneity of stock performance. Secondly, the model introduces industry average returns adjusted for annual volatility as hyperedge weights. This optimization improves the industry hypergraph structure and effectively reflects the risk level of each industry. Finally, a gating mechanism is integrated into the hypergraph convolution to filter and retain important features, further improving the precision and effectiveness of information propagation. Extensive experiments demonstrate that DAHL outperforms state-of-the-art models, achieving average return ratio of 69 %, 59 % and 63 % on China's A-share, National Association of Securities Dealers Automated Quotations (NASDAQ), and New York Stock Exchange (NYSE), respectively.