Leave or not leave? Group members' departure prediction in dynamic information networks (Q6071249)

This paper studies group members' departure prediction in dynamic information networks. It explores the problem of group members' departure prediction in dynamic information networks based on both groups' activeness and users' own activeness to improve the prediction precision. The group's activeness is defined considering both its structure features and members' activity information. Some techniques are proposed to efficiently compute each group's activeness. An unsupervised algorithm is presented to predict whether a user would leave a group with a sliding window model. Numerical experiments are performed on real-world dynamic information networks to show the effectiveness of the algorithm.