Mining Strongly Correlated Sub-graph Patterns by Considering Weight and Support Constraints

Frequent graph mining is one of famous data mining fields that receive the most attention, and its importance has been raised continually as recent databases in the real world become more complicated. Weighted frequent graph mining is an approach for applying importance of objects in the real world to the graph mining, and numerous studies related to this have been conducted so far. However, all of the results obtained from this approach do not become actually useful information, and a significant portion of them may be meaningless ones even though they are weighted frequent sub-graph patterns. To overcome this problem, in this paper, we propose a novel method which can consider whether any sub-graph pattern has close correlation among elements in the pattern, called MSCG (Mining Strongly Correlated sub-Graph). In experimental results, we demonstrate that our MSCG outperforms a state-ofthe-art method with respect to runtime and memory usage.