Redundant Operation Reduction Technique on Repeated Mining of Frequent Subgraphs

Frequent subgraphs represent intrinsic property of graphs and they can be used as significant features for various applications such as classification, clustering, and indexing of a graph DB. In order to select useful features, recent graph mining techniques applies repeated mining of frequent subgraphs either by varying minimum supports or by dividing a graph DB recursively. Such mining techniques suffer from long runtime, and most of the runtime is spent for the repeated mining of frequent subgraphs. In this paper, we discuss redundant execution of expensive canonical graph operations through analyzing the repeated mining of frequent subgraphs. We then propose a novel canonical graph search tree for indexing canonical graphs that can reduce redundant canonical graph operations for the same graph. We also propose a compression technique for the canonical graph search tree in order to reduce the maintenance cost of the tree. In experiment, we show that the proposed technique can reduce runtime by up to 15% compared with the existing model based search tree.