A Pattern Growth Approach for Frequent Subgraph Mining

Abstract

Graph mining has wide applications in chemistry, biology and computer networks. This kind of structure pattern mining might encounter more duplication cases due to graph isomorphism. Pattern growth approach has been shown to perform well for unstructured pattern mining, such as itemset and sequential patterns. In this paper, we shall examine whether such approach can also works for structured pattern mining. We propose a graph mining algorithm which enumerates frequent patterns by combining known frequent patterns with local frequent edges discovered via embedding lists. The embedding list technique not only facilitates the discovery of local frequent edges but also averts subgraph isomorphism checking. The empirical study on synthetic and real datasets demonstrates that HybridGMiner outperforms the algorithm gSpan but seconds to Gaston.