Triangle-Based Tripartite Graph Clustering

Abstract

A tripartite graph is three bipartite graphs between three vertex classes. In applications, it is three binary relationships between three sets of entities. The abundance of triangles in an otherwise sparse tripartite graph indicates that it is far from random and there are dense spots or clusters worth illuminating. Triangles in a tripartite graph form a graph by sharing edges. This triangle graph has some specific properties that make its clustering quite different from general graph clustering. In this work, we describe mode seeking in the triangle graph, based on a number of reasonable local density measures. Many of the properties and strategies for tripartite graphs described in this work extend to k-partite graphs including bipartite graphs.