ODU COMPUTATIONAL EXPERTS AIM TO IMPROVE INTERNET SEARCHES WITH NSF GRANT
A unique, collaborative classification system proposed by three Old Dominion University researchers to improve Internet searches of nontextual content has won $403,000 in developmental funds from the National Science Foundation.
Harris Wu, assistant professor of information technology and decision sciences in the College of Business and Public Administration, is principal investigator for the three-year grant. Working with him will be two computer science professors in the College of Sciences, Kurt Maly, an eminent scholar, and Mohammad Zubair.
Under the grant, the focus of the classification system will be documents and other items in the U.S. government's photograph and multimedia collection. Content that is not text-based is difficult to classify-or tag with search-friendly descriptive words. But the ODU team is devising a system that allows anyone who searches within the government collection to contribute key words that will enable the classification system to evolve and stay up-to-date.
This scheme is likened by the researchers to social tagging systems utilized by flickr.com and wiki classification interfaces. Users suggest keywords, and by their choices they can also accept, reject or modify the "soft" preliminary classifications that originate with the system.
Wu, Maly and Zubair recently performed a prototype implementation on an image subset of the government's multimedia collection titled "American Political History." This subset, which contains digital images of some of the nation's most valuable historical documents, "lacks tools and metadata for users to explore and utilize," according to a paper written by the three researchers.
"After deploying our prototype, the collection now can be browsed through several facts (and) users can create more facets as needed," they wrote. Also, the prototype will allow browsers to add structured metadata in a collaborative fashion by clicking on a "classify" button beside each entry.
Because they believe that user efforts alone cannot adequately classify documents, Wu offers an automated-algorithmic-control approach that imposes global order over the overlapping hierarchical classifications manually created by users.
The researchers say their evaluation so far suggests two major areas for them to work on under the grant: to prune the redundant categories created by users and to improve the accuracy of automated classification.
The grant, titled "Collaborative Classification of Large, Growing Collections with Evolving Facets," extends through August 2010.