Ranking people based on metadata analysis of search results

Abstract

Person search is one of the most popular search types on the Web. Most of the conventional technologies for person search focused on mapping the person name to a specific person (i.e. referents). In contrast, in this paper, we propose a novel ranking measure called famousness for person search. We use the notion of famousness for ranking people according to how well-known they are. Intuitively, famousness score is computed by analyzing the metadata of search results returned by a search engine. The metadata used in our method include URL, snippet, and the number of search results. To compute the famousness score of a person, first, we cluster the search results by using their metadata. Second, we compute the deviations in the size and number of such clusters. If the related Web pages of a person can be grouped into many large clusters of similar size, it looks like that person has been mentioned in many Web pages from various domains and that s/he is well known. In addition, we compare the clusters of search results with those of other people. Persons having more and larger clusters are given higher famousness scores. We also show experimental results to validate the ranking method based on the famousness score. © Springer-Verlag Berlin Heidelberg 2008.