Journals Proceedings

Abstract

We introduce a truss decomposition algorithm for bipartite graphs. A subgraph G of a graph is called k-truss if there are at least k-2 triangles containing any edge e of G. By a standard breadth-first-search algorithm, we can compute the truss decomposition for large graphs. To extract a dense substructure that represents community in graph G, this method avoids the intractable problem of clique detection. The truss decomposition is not, however, applicable to the bipartite graphs due to its definition. For this problem, we have proposed quasi-truss decomposition introducing an additional set of edges. For this decomposition, there is another problem such that dense subgraphs G1 and G2 are connected with a small number of edges. The previous algorithm detects the sparse structure H = G1 ⋃ G2 as quasi-truss due to the definition. In this paper, we improve the algorithm to extract denser substructures by removing such sparse edges with a top-down strategy. The extended algorithm has been implemented, and compared its performance with the previous algorithm for bipartite graphs obtained from real data.