EXTENDING THE DEWEY DECIMAL CLASSIFICATION VIA KEYWORD CLUSTERING - THE SCIENCE LIBRARY CATALOG PROJECT

The Science Library Catalog is a direct-manipulation browsing-oriented online catalog intended for use by children. Our research goals are to construct a system that provides an innovative interface tailored to children's cognitive development, interests and capabilities, yet can be implemented by loading standard MARC records from extant library collections. We have addressed these goals by reorganizing the catalog databases of individual libraries based on the Dewey Decimal Classification, creating a radically different visualization of the bibliographic data from that of standard online catalogs. We report here on the clustering algorithms used to extend the Dewey class number assignments on a large database of more than 8200 bibliographic records from a catalog developed over a century-long period. By combining clustering algorithms with the assistance of a human editor, we reassigned 7076 records from classification numbers containing an average of 126 items each (maximum 1140 records in one classification) into 959 clusters averaging 7.4 items each, producing a manageable retrieval hierarchy. Our results suggest that considerable information retrieval improvements can be made on extant bibliographic data without the expense of re-cataloging and reclassifying established collections.