EasyHypergraph: an open-source software for fast and memory-saving analysis and learning of higher-order networks

• Published in: Humanities & social sciences communications Vol. 12; no. 1; pp. 1291 - 19
• Main Authors: Ye, Bodian, Gao, Min, Zhan, Xiu-Xiu, He, Xinlei, Zhang, Zi-Ke, Gong, Qingyuan, Wang, Xin, Chen, Yang
• Format: Journal Article
• Language: English
• Published: London Palgrave Macmillan UK 09.08.2025; Springer Nature B.V; Springer Nature
• Subjects: 4007/2801; 4014/2801; Algorithms; Collaboration; Datasets; Flexibility; Graphs; Humanities and Social Sciences; Libraries; multidisciplinary; Open source software; Python; Researchers; Science; Science (multidisciplinary)
• ISSN: 2662-9992; 2662-9992
• DOI: 10.1057/s41599-025-05180-5

Higher-order relationships exist widely across different disciplines. In the realm of real-world systems, significant interactions involving multiple entities are common. The traditional pairwise modeling approach leads to the loss of important higher-order structures, while hypergraph is one of the most typical representations of higher-order relationships. To deeply explore the higher-order relationships, researchers and practitioners use hypergraph analysis to model the higher-order relationships and describe the important topological features in higher-order networks. At the same time, they carry out hypergraph learning studies to learn better node representations by designing hypergraph neural network models. However, existing hypergraph libraries still have the following research gaps. The first is that most of them are not able to support both hypergraph analysis and hypergraph learning, which negatively impacts the user experience. The second is that the existing libraries exhibit insufficient computational performance, which causes researchers and practitioners to spend more time and incur expensive resource costs. To fill these research gaps, we present EasyHypergraph, a comprehensive, computationally efficient, and storage-saving hypergraph computational library. To ensure comprehensiveness, EasyHypergraph designs data structures to support both hypergraph analysis and hypergraph learning. To ensure fast computation and efficient memory utilization, EasyHypergraph designs the computational workflow and demonstrates its effectiveness. Through experiments on five typical hypergraph datasets, EasyHypergraph saves at most 8470 s and 935 s over two baseline libraries in terms of analyzing node distance on a dataset with more than one hundred thousand nodes. For hypergraph learning, EasyHypergraph reduces HGNN training time by approximately 70.37% in a similar scenario. Finally, by conducting case studies for hypergraph analysis and learning, EasyHypergraph exhibits its usefulness in social science research.