Efficient traversal for core maintenance in large-scale dynamic hypergraphs

Yu Liu, Qi Luo, Yanwei Zheng, Zhipeng Cai, Dongxiao Yu

Abstract: The k-core in hypergraphs, where each vertex is incident to at least k hyperedges, is a critical cohesive subgraph model used in applications such as community detection, network robustness analysis, and information dissemination. However, existing methods for computing k-cores in dynamic hypergraphs face significant efficiency challenges due to time-consuming batch processing and redundant hyperedge traversals. To address these challenges, we develop a suite of efficient parallel core maintenance algorithms designed to enhance the speed and scalability of k-core computing in large-scale dynamic hypergraphs. Our algorithms rely on two key techniques: a batch processing structure called the Simultaneous Processing Set (SPS) and an advanced traversal strategy with pruning. SPS enables parallel processing of multiple hyperedges by grouping them based on disjoint key vertex sets, allowing simultaneous core number updates across independent regions of the hypergraph. This parallelization reduces redundant computations from repeatedly traversing the same regions during consecutive updates, improving computing efficiency. In addition, we introduce an advanced traversal strategy with pruning, which identifies and excludes hyperedges that do not affect core numbers. This is achieved by applying criteria based on incident hyperedges and core number thresholds for both vertices and hyperedges, eliminating irrelevant hyperedges from consideration. Extensive experiments demonstrate that the SPS batch management and advanced traversal strategy greatly enhance both computation speed and scalability. Our parallel algorithms achieve up to a 3x speedup compared to state-of-the-art multi-hyperedge algorithms, and offer efficiency improvements of up to 4 orders of magnitude when compared to single-edge techniques. Furthermore, our algorithms show excellent scalability as hypergraph sizes increase, making them highly effective for large-scale applications.

External IDs:dblp:journals/www/LiuLZCY25