A Model-based K-means Algorithm for Name Disambiguation

Unambiguous identities of resources are important aspect for semantic web. This paper addresses the personal identity issue in the context of bibliographies. Because of abbreviations or misspelling of names in publications or bibliographies, an author may have multiple names and multiple authors may share the same name. Such name ambiguity affects the performance of identity matching, document retrieval and database federation, and causes improper attribution of research credit. This paper describes a new K-means clustering algorithm based on an extensible Naïve Bayes probability model to disambiguate authors with the same first name initial and last name in the bibliographies and proposes a canonical name. The model captures three types of bibliographic information: coauthor names, the title of the paper and the title of the journal or proceeding. The algorithm achieves best accuracies of 70.1% and 73.6% on disambiguating 6 different “J Anderson” s and 9 different "J Smith" s based on the citations collected from researchers’ publication web pages.