Families of Butterfly Counting Algorithms for Bipartite Graphs

Abstract

Butterflies are an important motif found in bipartite graphs that provide a structural way for finding dense regions within the graph. Beyond counting butterflies and enumerating them, other metrics and peeling for bipartite graphs are designed around counting butterfly motifs. The importance of counting butterflies has led to many works on efficient implementations for butterfly counting, given certain situational or hardware constraints. However, most algorithms are based on first counting the building block of the butterfly motif, and from that calculating the total possible number of butterflies in the graph. In this paper, using a linear algebra approach, we show that many provably correct algorithms for counting butterflies can be systematically derived. Moreover, we show how this formulation facilitates butterfly peeling algorithms that find the k-tip and k-wing subgraphs within a bipartite graph.