| The focus of this thesis is to apply relational database techniques to support frequent subgraph mining over a set of graphs. Our primary goal is to address scalability of graph mining to very large data sets, not currently addressed by main memory approaches. Unlike the main memory counter parts, this thesis addresses the most general graph representation including multiple edges between any two vertices, and cycles. In the process of developing frequent subgraph mining over a set of graphs, the substructure representation of HDB-Subdue has been leveraged and extended. An algorithm is presented for frequent subgraph mining over a set of graphs. We also present an algorithm for pseudo duplicate elimination that is more efficient than the one used in the previous approach (HDB-Subdue).;This thesis also presents an efficient approach to infer structural relationships from relational data to facilitate graph mining (either the best subgraph or frequent subgraphs). (Abstract shortened by UMI.)... |
