Correlated subgraph search for multiple query graphs in graph streams

Abstract

In real-world, there are many dynamic graph databases. Correlation mining is one of important analysis method in these dynamic graph databases. CGStream has been proposed to discover correlated graphs efficiently in graph streams. However, CGStream suffers from long running time for multiple queries. CGStream must perform frequent subgraph mining n times for n queries to generate candidate patterns. Many of the candidate patterns are redundant because multiple queries can have the same candidates. In this paper, we propose an efficient framework, MCGStream that supports correlated subgraph search for multiple queries in graph streams. The proposed framework generates no redundant candidate pattern for multiple queries by performing frequent subgraph mining with the global lower frequency bound. We propose the method that determines the global lower frequency bound for multiple queries. We build a correlation candidate tree to maintain the candidate patterns and their mapping to queries. Several optimization techniques are applied to the correlation candidate tree to reduce the search space. Experiments show that the proposed method efficiently processes multiple queries compared with CGStream by up to 35%.