CICN: Higher-order link prediction with clustering mutual information of common neighbors

Abstract

In complex networks, traditional link prediction focuses on pairwise interactions between node pairs to predict missing links and identify spurious interactions, which has a wide range of applications in the real world. However, with the continuous expansion of network scale, interactions within the network occur not only between pairs of nodes but also involve higher-order interactions among multiple nodes. However, traditional link prediction methods face challenges in directly predicting these higher-order structures. In this paper, we propose a novel higher-order link prediction method based on Clustering mutual Information of Common Neighbors (CICN) for the prediction of 3-cliques (triangles). CICN employs node information entropy and calculates the impact of common neighbors by integrating different-order clustering coefficients to predict 3-cliques in the network. Experiments on 9 empirical networks show that the higher-order clustering patterns of nodes can significantly improve the accuracy of predicting 3-cliques. Additionally, we investigate the stability of the proposed algorithm and the results indicate that the performance of the CICN remains favorable across training sets of varying sizes. The source codes of our method are publicly available at: https://github.com/yabingyao/CICN4HigherOrderLinkPrediction.