Affiliations:
College of Computer, Nanjing University of Posts and Telecommunications, Nanjing, China
Correspondence:
[*]
Corresponding author. Ke-Jia Chen, College of Computer, Nanjing University of Posts and Telecommunications, Nanjing 210046, China. Tel./Fax: +86 18912963366; E-mail: chenkj@njupt.edu.cn..
Note: [1] This work was supported by the National Nature Science Foundation of China (No. 61100135, 61302157, 61302158).
Abstract: Link prediction is an important sub-task in link mining area. This paper discusses link prediction in dynamic networks and proposes a new link prediction method which can learn from the long-term graph evolution of networks. The method first represents the variation of the structural properties in a dynamic network. Then, a classifier is trained for each property. It finally conducts link prediction process using an ensemble result of all the classifiers. Experiments in three realistic collaboration networks show that the evolution information of the network is beneficial for the improvement of link prediction performance and different structural property has different capability to describe dynamics of the network.
Keywords: Dynamic network, link prediction, machine learning, social network analysis
