Adaptive clustering of scientific data

Abstract

Scientific databases contain large amounts of interrelated information. This information is often stored in relational databases with hundreds of tables and thousands of rows per table. Clustering is an effective way to reduce the information-overhead associated with finding information among these tables, allowing the user to browse through the clusters as well as the individual tables. In this paper, we compare the use of two adaptive algorithms (genetic algorithms, and neural networks) in clustering the tables of a scientific database. These clusters allow the user to index into this overwhelming number of tables and find the needed information quickly. We cluster the tables based on the user’s queries and not on the content of the tables, thus the clustering reflects the unique relationships each user sees among the tables. The original database remains untouched, however each user will now have a personalized index into this database.