Temporal-Relational hypergraph tri-Attention networks for stock trend prediction

•We introduce two heterogeneous hypergraphs to separately characterize the groupwise relationships of industry-belonging and fund-holding among stocks. To the best of our knowledge, we are the first to leverage both the group-wise relationships of industry-belonging and fund-holding relationships for stock trend prediction.•We propose a novel hypergraph tri-attention network (HGTAN) that consists of hierarchical attention modules to consider the importance of different nodes, hyperedges, and hypergraphs when guiding the information propagation in stock hypergraphs.•We conduct both experimental evaluation and investment simulation on real-world data, and the results demonstrate the validity and rationality of our approach.