Irene Eleta
Ph.D. Candidate
University of Maryland
Raul Guerra
Ph.D. Candidate
University of Maryland
Museums and libraries have growing collections of digital images of their art works. Traditionally to enable access, experts create authoritative metadata for these images. About five years ago, the Institute of Museum and Library Services (IMLS)-funded steve.museum project explored the use of social tagging by non-experts to create image labels. At about the same time, the Computational Linguistics for Metadata Building (CLiMB) project, funded by the Mellon Foundation, explored ways to extract terms from text on those images. The results of steve.museum along with CLiMB enabled researchers at the University of Maryland, the Indianapolis Museum of Art, and Susan Chun, consultant, to consider how to combine these valuable sources for access.
This session will include a presentation of the results of this IMLS-funded research, the Text, Tags and Trust (T3) Project. The presentation will focus on two fundamental issues in using user-created metadata. The computational linguistic processing, morphological and semantic analysis techniques used to process and analyze the large steve.museum tagset will be discussed first. Tags and terms will be compared to show how each covers different types of issues and vocabularies. While tagging tends to be by the non-expert, terms from text may be more authoritative, but the volume of irrelevant phrases impacts overall usefulness. The computational techniques used in this project enabled the examination of tags and phrases of importance for browsing, and discriminate useful multi-word phrases with high descriptive value.
Results of a comparison of social tagging patterns in two languages will also be presented, and exploitable strengths for providing multilingual support in digital libraries and museums will be explored. Additional text metadata could be leveraged for effective image browsing, e.g. by reducing noise, filtering of results, suggesting terms, recommending images, and clustering of these images for browsing.
Project contributors include Judith Klavans, Jen Golbeck, Susan Chun, Rob Stein, Ed Bachta, Irene Eleta, Raul Guerra, and Rebecca LaPlante.
http://umiacs.umd.edu/research/t3/
Handout (PDF)