With ever-growing storage needs and drift towards very large relational storage settings, multi-relational data mining has become a prominent and pertinent field for discovering unique and interesting relational patterns. As a consequence, a whole suite of multi-relational data mining techniques is being developed. These techniques may either be extensions to the already existing single-table mining techniques or may be developed from scratch. For the traditionalists, single-table mining algorithms can be used to work on multi-relational settings by making inelegant and time consuming joins of all target relations. However, complex relational patterns cannot be expressed in a single-table format and thus, cannot be discovered. This work presents a new multi-relational frequent pattern mining algorithm termed Multi-Relational Frequent Pattern Growth (MRFP Growth). MRFP Growth is capable of mining multiple relations, linked with referential integrity, for frequent patterns that satisfy a user specified support threshold. Empirical results on MRFP Growth performance and its comparison with the state-of-the-art multirelational data mining algorithms like WARMR and Decentralized Apriori are discussed at length. MRFP Growth scores over the latter two techniques in number of patterns generated and speed. The realm of multi-relational clustering is also explored in this thesis. A multi-Relational Item Clustering approach based on Hypergraphs (RICH) is proposed. Experimentally RICH combined with MRFP Growth proves to be a competitive approach for clustering multi-relational data. The performance and iii quality of clusters generated by RICH are compared with other clustering algorithms. Finally, the thesis demonstrates the applied utility of the theoretical implications of the above mentioned algorithms in an application framework for auto-annotation of images in an image database. The system is called CoMMA which stands for Combining Multi-relational Multimedia for Associations.
Library of Congress Subject Headings
Relational databases; Data mining; Data structures (Computer science); Pattern perception; Cluster analysis
Department, Program, or Center
Computer Science (GCCIS)
Teredesai, Ankur - Chair
Kanodia, Juveria, "Structural advances for pattern discovery in multi-relational databases" (2005). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus