Incorporating user grouping into retweeting behavior modeling

Abstract

The variety among massive users makes it difficult to model their retweeting activities. Obviously, it is not suitable to cover the overall users by a single model. Meanwhile, building one model per user is not practical. To this end, this paper presents a novel solution, of which the principle is to model the retweeting behavior over user groups. Our system, GruBa, consists of three key components for extracting user based features, clustering users into groups, and modeling upon each group. Particularly, we look into the user interest from different perspectives including long-term/short-term interests and explicit/implicit interests. We have evaluated the performance of GruBa using datasets of real-world social networking applications, showcasing its benefits.