Learning to rank domain experts in microblogging by combining text and non-text features

Currently microblog search engines have the function to find related users according to input topic keywords. Traditional approaches rank users by their authentication information or their self descriptions (introductions or labels).However, many users may not publish the posts closely related to their certification profile. In this paper, we study the problem of identifying domain-dependent influential users (or topic experts). We propose to fuse of non-text features and text features to analysis the influence of the users. In addition we compare three kinds of sorting methods, i.e., order-based rank aggregation, greedy selection based rank aggregation, SVM Rank method. Our experimental results show that the highest precision is achieved by SVM rank method.