An Experiment with Link Prediction in Social Network: Two New Link Prediction Methods

Abstract

This paper investigates link prediction methods in social networks (Facebook) and discusses two new link prediction methods. The two methods are structural based methods, which implies they are not based on any content of user profiles, but instead they are based on connections of the users in the social network. These two methods are the Common Neighbors of Neighbors and Node Connectivity prediction methods. The first introduced method can be considered as an extension to the Common Neighbors link prediction method. The second method is based on average connections of neighbors. Both methods are discussed in this paper and have been used in experiment. Additionally, Formulas, explanation, Pseudocode and an example is included about some preexisting methods of link prediction and the introduced methods considered in this paper. This paper also includes detailed applications of link predictions methods, and experimental results that compare the new methods with well-known methods in link prediction. Results show better performance for the proposed link prediction method when applied to a friendship Facebook dataset, which is characterized in this paper. The experiment is described in details and the results (precision and number of positives) shows superiority of proposed methods in terms of performance over well-known link prediction methods.