Top-K Correlation Sub-graph Search in Graph Databases

Recently, due to its wide applications, (similar) subgraph search has attracted a lot of attentions from database and data mining community, such as [13, 18, 19, 5]. In [8], Ke et al. first proposed correlation sub-graph search problem (CGSearch for short) to capture the underlying dependency between sub-graphs in a graph database, that is CGS algorithm. However, CGS algorithm requires the specification of a minimum correlation threshold theta to perform computation. In practice, it may not be trivial for users to provide an appropriate threshold theta, since different graph databases typically have different characteristics. Therefore, we propose an alternative mining task: top-K correlation sub-graph search (TOP-CGSearh for short). The new problem itself does not require setting a correlation threshold, which leads the previous proposed CGS algorithm inefficient if we apply it directly to TOP-CGSearch problem. To conduct TOP-CGSearch efficiently, we develop a pattern-growth algorithm (that is PG-search algorithm) and utilize graph indexing methods to speed up the mining task. Extensive experiment results evaluate the efficiency of our methods.