Efficient Structural Clustering in Large Uncertain Graphs

Abstract

Clustering uncertain graphs based on the probabilistic graph model has sparked extensive research and widely varying applications. Existing structural clustering methods rely heavily on the computation of pairwise reliable structural similarity between vertices, which has proven to be extremely costly, especially in large uncertain graphs. In this paper, we develop a new, decomposition-based method, ProbSCAN, for efficient reliable structural similarity computation with theoretically improved complexity. We further design a cost-effective index structure UCNO-Index, and a series of powerful pruning strategies to expedite reliable structural similarity computation in uncertain graphs. Experimental studies on eight real-world uncertain graphs demonstrate the effectiveness of our proposed solutions, which achieves orders of magnitude improvement of clustering efficiency, compared with the state-of-the-art structural clustering methods in large uncertain graphs.