Accurate link prediction method based on path length between a pair of unlinked nodes and their degree.

The link prediction problem has received much attention since the beginnings of social and behavioral sciences. For instance, social networks such as Facebook, Twitter, and LinkendIN change enduringly as new connections appear in the graph. For these networks, one of the biggest challenges is to find accurately the best recommendations to the users. Within the meaning of the graph, the main objective of the link prediction problem is to predict the upcoming links from the actual state of a graph. Link prediction methods use some score functions, such as Jaccard coefficient, Katz index, and Adamic Adar metric, to measure the probability of adding the links to the network. These metrics are widely used in various applications due to their simplicity and their interpretability; however, the majority of them are designed for a specific domain. Social networks become very large with a several number of users that are connected with different kinds of links. Predicting those links is still a challenging task, as we need to find the best way to perform predictions as accurate as possible. Along this way, we extend our previous work is (Jibouni et al. in 2018 6th international conference on wireless networks and mobile communications (WINCOM). IEEE, pp 1–6, 2018) where we have proposed a new node similarity measure based on the path depth between the source and destination nodes and their degrees. The used topological features are very easy to compute and very effective in solving the link prediction problem. In addition, we verify the impact of the path length l on the method performance and we show that the proposed method provides more accurate recommendations by using the path length 2 and 3. Then, we compare 13 state-of-the-art methods against the proposed method in terms of their prediction performance using the area under curve. The results on five instances of social networks show the efficiency of the proposed method in providing accurate recommendations. Furthermore, we consider machine learning techniques such as K-nearest neighbors, logistic regression, artificial neural network, decision tree, random forest, support vector machine to solve the link prediction problem as a binary classification task. The results confirm the significant accuracy improvement that can be achieved using the proposed metric.