Learning to Find Topic Experts in Twitter via Different Relations

Abstract

Expert finding has become a hot topic along with the flourishing of social networks, such as micro-blogging services like Twitter. Finding experts in <i>Twitter</i> is an important problem because tweets from experts are valuable sources that carry rich information (e.g., trends) in various domains. However, previous methods cannot be directly applied to <i>Twitter</i> expert finding problem. Recently, several attempts use the relations among users and <i>Twitter List</i> s for expert finding. Nevertheless, these approaches only partially utilize such relations. To this end, we develop a probabilistic method to jointly exploit three types of relations (i.e., <i>follower</i> relation, <i>user-list</i> relation, and <i>list-list</i> relation) for finding experts. Specifically, we propose a <i>S</i> emi- <i>S</i> upervised <i>G</i> raph-based <i>R</i> anking approach ( <inline-formula><tex-math notation="LaTeX">$\sf{SSGR}$</tex-math></inline-formula> ) to offline calculate the <i>global authority</i> of users. In <inline-formula><tex-math notation="LaTeX">$\sf{SSGR}$</tex-math></inline-formula> , we employ a normalized Laplacian regularization term to jointly explore the three relations, which is subject to the supervised information derived from Twitter crowds. We then online compute the <i>local relevance</i> between users and the given query. By leveraging the <i>global authority</i> and <i>local relevance</i> of users, we rank all of users and find top-N users with highest ranking scores. Experiments on real-world data demonstrate the effectiveness of our proposed approach for <i>topic-specific</i> expert finding in <i>Twitter</i> .