Mining the URLs: An Approach to Measure the Similarities between Named-Entities

Measuring the similarity between named-entities is a foundation work for a number of practical applications, such as information extraction, query expansion, etc. In this paper the authors study the similarity measure between two named-entities. Especially, the authors are interested in fine-grained similarity differences between named-entities in one class, such as "novelist". Different from previous works on named-entity associations, this paper suggests a novel Web mining method that solely depends on the URLs returned by a search engine using named-entities as queries. The problem of similarity between two namedentities is converted to that of similarity of two URL sets. Evaluations show that this method achieves good results under two experiments.