Structural correlation pattern mining for large graphs

Arlei Silva, Wagner Meira Jr., Mohammed J. Zaki

Abstract: In this paper we define the Structural Correlation Pattern (SCP) mining problem, which consists of determining correlations among vertex attributes and dense components in an undirected graph. Vertex attributes play an important role in several real-life graphs and SCPs help to understand how they relate to the associated graph topology. SCPs may describe, for example, interesting relationships between personal characteristics and the community structure in social networks. We also propose an efficient algorithm, called SCORP, to extract SCPs from large graphs, and compare it against a naive approach for SCP mining, demonstrating its scalability and efficiency. We also discuss the application of SCORP to two actual scenarios, co-authorship networks and social music discovery, showing relevant results that demonstrate the applicability of the proposed approach.