Role Centralized Modeling for Expert Search in Enterprise Corporation

Automatic expert finding systems aiming at identifying experts from a large set of document repository have attracted considerable interest in recent years. To better describe the relationship on expert-document and document-topic, we introduce the conception “Role” in our model and expand its use in expert search. We illustrate how the conception “Role” in our model helps to present an effective approach to model the probability of the candidate being an expert on the given topic with documents relevant to it. Consequently, we evaluate the effectiveness of our model by a series of experiments on TREC 2006 Enterprise Track expert search task. The results show that our role centralized model performs much better than general approaches overall. It is a big potential to dig in more deeply in the future.