Finding Clusters in Network Link Strength Data

In this paper we introduce BEACON, a tool for nding clusters of objects using data which are measurements of strengths of links between those objects. This technique is useful for measuring modularity in software systems as well as in analysis of social network data. We apply Bayesian tools such as Markov chain Monte Carlo to estimate clusters. The paper discusses simulation experiments which demonstrate the power of the methodology to nd true clusters, and applies the technique to the analysis of a subsystem of a large telecommunications software product.