An approach for predicting missing links in social network using node attribute and path information

In social networks, link prediction is the task to identify links in future. Many existing link prediction techniques used similarity scores to predict links. An essential concern in the link prediction problem is identifying missing links between the nodes when there are no common neighbors between the nodes. Considering this, a new algorithm proposed, namely Similarity-based Algorithm using Degree and Common Neighbour (SADCN) which includes a node's degree in the shortest path and common neighbor. For experiment evaluation, three datasets are used to test our method performance against some standard similarity index and the recently proposed algorithms for link prediction, which depicts that our approach achieved comparable AUC values to those that consider common neighbors and it gives better AUC for those links, where no mutual neighbour between the two nodes exists. Finally, we create feature vectors and use XGB classifiers for predicting links. It shows that our proposed algorithm can improve the F-measure and accuracy in a feature based link prediction model.