DB-subdue: Database approach to graph mining

Abstract

In contrast to mining over transactional data, graph mining is done over structured data represented in the form of a graph. Data having structural relationships lends itself to graph mining. Subdue is one of the early main memory graph mining algorithms that detects the best substructure that compresses a graph using the minimum description length principle. Database approach to graph mining presented in this paper overcomes the problems – performance and scalability – inherent to main memory algorithms. The focus of this paper is the development of graph mining algorithms (specifically Subdue) using SQL and stored procedures in a Relational database environment. We have not only shown how the Subdue class of algorithms can be translated to SQL-based algorithms, but also demonstrated that scalability can be achieved without sacrificing performance.