A parallel algorithm for frequent subgraph mining

Abstract

Graph mining has practical applications in many areas such as molecular substructure explorer, web link analysis, fraud detection, outlier detection, chemical molecules, and social networks. Frequent subgraph mining is an important topic of graph mining. The mining process is to find all frequent subgraphs over a collection of graphs. Numerous algorithms for mining frequent subgraphs have been proposed; most of them, however, used sequential strategies which are not scalable on large datasets. In this paper, we propose a parallel algorithm to overcome this weakness. Firstly, the multi-core processor architecture is introduced; the way to apply it to data mining is also discussed. Secondly, we present the gSpan algorithm as the basic framework of our algorithm. Finally, we develop an efficient algorithm for mining frequent subgraphs relied on parallel computing. The performance and scalability of the proposed algorithm is illustrated through extensive experiments on two datasets, chemical and compound.