Similarity-Based Hybrid Algorithms for Link Prediction Problem in Social Networks

In this study, we propose two hybrid algorithms for link prediction problem in static networks that combine the benefits of both local and global scoring methods, with the objective of compensating the weaknesses of each approach using two strategies: sequential and integrated. In the sequential strategy global scoring methods are used in a pipeline mode after local ones if the full graph is explored and the desired number of edges is not met, in an attempt to complete the missing links in the network. The integrated one combines local and global scoring algorithms together in order to add a missing link to the network. Furthermore, we present four distinct approaches to explore the network’s nodes and edges. Experiments on real-world and synthetic networks revealed that our proposed hybrid algorithms can outperform some of the state-of-the-art link-prediction methods.