Weighted experts : a solution for the spock data mining challenge

One of the most popular and trend-setting Internet applications is People Search on the World Wide Web. In its most general form, information extraction for persons from unstructured data is extremely challenging, and, we are pretty far away from satisfying solutions. However, current retrieval technology is able to cope with restricted variants of the problem, and this paper deals with such a variant, the so-called multi document person resolution. Given is a set of Web documents, and the task is to state for each document pair whether the two documents are talking about the same person or not. For this problem Spock Inc., Silicon Valley, launched in 2007 a competition offering a grand prize of $50 000. Task was the person-specific classification of 100 000 Web pages within 4 hours on a standard PC, striving for a maximum F-Measure. The paper in hand describes the challenge and introduces the technology of the winning team from the Bauhaus University Weimar [see 1]. -- Person resolution ; named entity recognition ; supervised cluster analysis