Efficient and Effective Attributed Hypergraph Clustering via K-Nearest Neighbor Augmentation

Abstract

Hypergraphs are an omnipresent data structure used to represent high-order interactions among entities. Given a hypergraph H wherein nodes are associated with attributes, attributed hypergraph clustering (AHC) aims to partition the nodes in H into k disjoint clusters, such that intra-cluster nodes are closely connected and share similar attributes, while inter-cluster nodes are far apart and dissimilar. It is highly challenging to capture multi-hop connections via nodes or attributes on large attributed hypergraphs for accurate clustering. Existing AHC solutions suffer from issues of prohibitive computational costs, sub-par clustering quality, or both. In this paper, we present AHCKA, an efficient approach to AHC, which achieves state-of-the-art result quality via several algorithmic designs. Under the hood, AHCKA includes three key components: (i) a carefully-crafted K-nearest neighbor augmentation strategy for the optimized exploitation of attribute information on hypergraphs, (ii) a joint hypergraph random walk model to devise an effective optimization objective towards AHC, and (iii) a highly efficient solver with speedup techniques for the problem optimization. Extensive experiments, comparing AHCKA against 15 baselines over 8 real attributed hypergraphs, reveal that AHCKA is superior to existing competitors in terms of clustering quality, while often being up to orders of magnitude faster.