Optimized Candidate Generation for Frequent Subgraph Mining in a Single Graph

Abstract

Mining frequent subgraphs from graph databases is a basic task with broad applications. Frequent subgraph mining is defined as finding all subgraphs that appear more than specified threshold value. It consists of mainly two steps, candidate generation and frequency calculation. In candidate generation step, most of the existing work starts with a frequent edge or vertex to generate frequent candidate patterns. This process is not scalable due to exponential number of candidate patterns generation. In this paper, an optimized algorithm is presented to generate candidate patterns for mining frequent subgraphs from a large single graph. The proposed algorithm starts and extends candidates with frequent subgraphs. The proposed algorithm uses graph invariant properties and symmetries present in a graph to generate candidate subgraphs thus reducing generation of enormous amount of candidate subgraphs. Subgraphs are extended by adding another frequent subgraph determined by the symmetry mapping of subgraph there by reduces the complexities involved in candidate generation and frequency counting. An evaluation study on datasets explores the strengths and limitations of the proposed work. The results make sure that, this is an optimized approach to generate candidate subgraphs directly using invariant properties.