CloseGraph: Mining Closed Frequent Graph Patterns

Abstract

Recent research on pattern discovery has progressed form mining frequent <i>itemsets</i> and <i>sequences</i> to mining structured patterns including <i>trees, lattices</i>, and <i>graphs</i>. As a general data structure, <i>graph</i> can model complicated relations among data with wide applications in bioinformatics, Web exploration, and etc. However, mining large graph patterns in challenging due to the presence of an exponential number of frequent subgraphs. Instead of mining all the subgraphs, we propose to mine <i>closed frequent graph patterns</i>. A graph <i>g</i> is <i>closed</i> in a database if there exists no proper supergraph of <i>g</i> that has the same support as <i>g</i>. A closed graph pattern mining algorithm, CloseGraph, is developed by exploring several interesting pruning methods. Our performance study shows that CloseGraph not only dramatically reduces unnecessary subgraphs to be generated but also substantially increases the efficiency of mining, especially in the presence of large graph patterns.