Frequent Sub-graph Mining on Edge Weighted Graphs

Frequent sub-graph mining entails two significant overheads. The first is concerned with candidate set generation. The second with isomorphism checking. These are also issues with respect to other forms of frequent pattern mining but are exacerbated in the context of frequent sub-graph mining. To reduced the search space, and address these twin overheads, a weighted approach to sub-graph mining is proposed. However, a significant issue in weighted sub-graph mining is that the antimonotone property, typically used to control candidate set generation, no longer holds. This paper examines a number of edge weighting schemes; and suggests three strategies for controlling candidate set generation. The three strategies have been incorporated into weighted variations of gSpan: ATW-gSpan, AW-gSpan and UBW-gSpan respectively. A complete evaluation of all three approaches is presented.