Multi document person resolution tackles the problem that the mapping between names and people (the denoted individuals or "referents") is not one-to-one. When searching the World Wide Web for a person (a "referent"), one is immediately faced with this problem: search results contain web pages of different individuals having the same name. The problem is addressed by integrating semantic feature analysis with unsupervised classification technology to group those pages that belong to the same referent.
With our research we have contributed highly effective solutions to this problem; they combine state of the art retrieval models, multi-staged density-based clustering algorithms, and a new class of clustering validity measures. The effectiveness of our approach has been demonstrated by winning the Spock Data Mining Challenge awarded with USD 50,000 in 2008. [award]
Students: Steffen Becker, Christof Bräutigam, Tino Rüb, and Hagen-Christian Tönnies