Detecting Patterns of Interaction in Temporal Hypergraphs via Edge Clustering

Abstract

Finding densely connected subsets of vertices in an unsupervised setting, called clustering or community detection, is one of the fundamental problems in network science. The edge clustering approach instead detects communities by clustering the edges of the graph and then assigning a vertex to a community if it has at least one edge in that community, thereby allowing for overlapping clusters of vertices. We apply the idea behind edge clustering to temporal hypergraphs, an extension of a graph where a single edge can contain any number of vertices and each edge has a timestamp. Extending to hypergraphs allows for many different patterns of interaction between edges, and by defining a suitable structural similarity function, our edge clustering algorithm can find clusters of these patterns. We test the algorithm with three structural similarity functions on a large collaboration hypergraph, and find intuitive cluster structures that could prove useful for downstream tasks.

12 pages, 11 figures, 1 table

References in corpus (13)

• SciPy 1.0--Fundamental Algorithms for Scientific Computing in Python
• Community detection in graphs
• UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
• The structure of scientific collaboration networks
• Link communities reveal multiscale complexity in networks
• Networks beyond pairwise interactions: structure and dynamics
• Line Graphs, Link Partitions and Overlapping Communities
• Community Discovery in Dynamic Networks: a Survey
• Line Graphs of Weighted Networks for Overlapping Communities
• Clustering via Hypergraph Modularity
• The simpliciality of higher-order networks
• Improving Mapper's Robustness by Varying Resolution According to Lens-Space Density
• Improving community detection via community association strength scores