Efficient mining of minimal distinguishing subgraph patterns from graph databases

Abstract

Distinguishing patterns represent strong distinguishing knowledge and are very useful for constructing powerful, accurate and robust classifiers. The d istinguishing g raph p atterns(DGPs) are able to capture structure differences between any two categories of graph datasets. Whereas, few previous studies worked on the discovery of DGPs. In this paper, as the first, we study the problem of mining the complete set of minimal DGPs with any number of positive graphs, arbitrary positive support and negative support. We proposed a novel algorithm, MDGP-Mine, to discover the complete set of minimal DGPs. The empirical results show that MDGP-Mine is efficient and scalable. © 2008 Springer-Verlag Berlin Heidelberg.