Dhaka University Journal of Applied Science & Engineering

Frequent pattern mining is a core problem in data mining. Algorithms for frequent pattern mining have been proposed for itemsets, sequences, and graphs. However, existing graph mining frameworks follow an edge-growth approach to building patterns which limits many applications. Motivated by real-life problems, in this work, we define a novel graph mining framework that incorporates vertex-based extensions along with the edge-growth approach. We also propose an efficient algorithm for mining frequent subgraphs. To deal with the exploding search space, we introduce a canonical labeling technique for isomorphic candidates as well as downward closure property-based search space pruning. We present an experimental analysis of our algorithm on real-life benchmark graph datasets to demonstrate the performance in terms of runtime.

1. ref1