Missing link prediction using path and community information

Due to the evolving nature of complex networks, link prediction plays a crucial role in exploring likelihood of potential relationships among nodes. There exist a great number of strategies to apply the similarity-based metrics for estimating proximity of nodes in complex networks. In this paper, we propose three new variants – CCPAL3, LPCPA, and GPCPA – for the well-known Common Neighbor and Centrality-based Parameterized Algorithm (CCPA) taking into account 3-hop path, quasi-local path, and global path, respectively. In addition, four novel link prediction strategies based on community detection information, CCPA_CD, CCPAL3_CD, LPCPA_CD and GPCPA_CD, are proposed. Meanwhile, the Jaccard index is extended to three new metrics, i.e., Jaccard_L3, Jaccard_QuasiLoc and Jaccard_Global. Extensive experiments are conducted on thirteen real-world networks. The experimental results indicate that the proposed metrics improve the prediction accuracy measured by AUC and are more competitive on Precision compared to the state-of-the-art link prediction methods.