Learning to Diversify Expert Finding with Subtopics

Expert finding is concerned about finding persons who are knowledgeable on a given topic. It has many applications in enterprise search, social networks, and collaborative management. In this paper, we study the problem of diversification for expert finding. Specifically, employing an academic social network as the basis for our experiments, we aim to answer the following question: Given a query and an academic social network, how to diversify the ranking list, so that it captures the whole spectrum of relevant authors’ expertise? We precisely define the problem and propose a new objective function by explicitly incorporating topic-based diversity into the relevance ranking measurement. A learning-based model is presented to solve the objective function. Our empirical study on a real system validates the effectiveness of the proposed method. The proposed method can achieve significant improvements (+15.3%-+94.6% by MAP) over alternative methods.