WFSM-MaxPWS: An Efficient Approach for Mining Weighted Frequent Subgraphs from Edge-Weighted Graph Databases

Weighted frequent subgraph mining comes with an inherent challenge—namely, weighted support does not support the downward closure property, which is often used in mining algorithms for reducing the search space. Although this challenge attracted attention from several researchers, most existing works in this field use either affinity based pruning or alternative anti-monotonic weighting technique for subgraphs other than average edge-weight. In this paper, we propose an efficient weighted frequent subgraph mining algorithm called WFSM-MaxPWS. Our algorithm uses the MaxPWS pruning technique, which significantly reduces search space without changing subgraph weighting scheme while ensuring completeness. Our evaluation results on three different graph datasets with two different weight distributions (normal and negative exponential) showed that our WFSM-MaxPWS algorithm led to significant runtime improvement over the existing MaxW pruning technique (which is a concept for weighted pattern mining in computing subgraph weight by taking average of edge weights).