Concept Disambiguation for Improved Subject Access Using Multiple Knowledge Sources

We address the problem of mining text for relevant image metadata. Our work is situ- ated in the art and architecture domain, where highly specialized technical vocabu- lary presents challenges for NLP tech- niques. To extract high quality metadata, the problem of word sense disambiguation must be addressed in order to avoid leading the searcher to the wrong image as a result of ambiguous — and thus faulty — meta- data. In this paper, we present a disam- biguation algorithm that attempts to select the correct sense of nouns in textual de- scriptions of art objects, with respect to a rich domain-specific thesaurus, the Art and Architecture Thesaurus (AAT). We per- formed a series of intrinsic evaluations us- ing a data set of 600 subject terms ex- tracted from an online National Gallery of Art (NGA) collection of images and text. Our results showed that the use of external knowledge sources shows an improvement over a baseline.