Abstract
When a variety of names are used for the same real-world entity, the problem of detecting all such variants has been known as the (record) linkage or entity resolution problem. In this paper, toward this problem, we propose a novel approach that uses the Web as the collective knowledge source in addition to contents of entities. Our hypothesis is that if an entity e1 is a duplicate of another entity e2, and if e1 frequently appears together with information I on the Web, then e2 may appear frequently with I on the Web. By using search engines, we analyze the frequency, URLs, or contents of the returned web pages to capture the information I of an entity. Extensive experiments verify that our hypothesis holds in many real settings, and the idea of using the Web as the additional source for the linkage problem is promising. Our proposal shows 51% (on average) and 193% (at best) improvement in precision/recall compared to a baseline approach.
Min-Yen Kan
Associate Professor
WING lead; interests include Digital Libraries, Information Retrieval and Natural Language Processing.