Visualizing Similarity in Subject Term Co‐Assignment
Jeff Gabel · Advances in Classification Research Online · 2009
The purpose of this research is to improve retrieval performance in systems that use assigned subject descriptors, such as library subject headings.We are looking for wider semantic boundaries surrounding summary headings assigned to documents by providing a means of identifying clustered headings that fall within the indexer's collective common perceptions of relevance.We are here experimenting with two techniques that can help increase both precision and recall.In earlier research citation-chasing was employed to yield a fuller retrieval set than might have been found using subject headings alone.In the present study we are employing multi-dimensional scaling to determine the best fit among works to which subject descriptors have been co-assigned.A term co-occurrence matrix compiled from 19 LCSH subject headings assigned to works in the field of "language origin" is used to generate an MDS map of the semantic space.Two clusters emerge: language and languages, and evolution biology, sometimes termed evolingo.Results allow us to visualize how differing perceptions of indexers affect the semantic space surrounding assigned terms.In both cases-citationchasing and term co-occurrence-and especially when combining the two techniques acting as thresholds for each other, it is possible to overcome the inverse relation between precision and recall.