The Adaptive Similarity-based Method For Link Prediction

In the past decade,the studies on complex networks are rapidly expanded and draw increasing attention from different disciplines.As one of the most important direction of complex networks,link prediction help researchers explore network structure,and explain the mechanism of network evolution.The similarity-based method is the simplest framework of link prediction,and has become the mainstream issue.However,existing similarity-based methods assume that similar nodes may be connect with each other,which lead to poor performance when clustering coefficient is not large enough.Different from traditional similarity-based methods only focus on pair-wise relationships between nodes,we utilize neighboring nodes to describe each node,and combine similarity and link prediction into a single framework,then propose an adaptive similarity regularized method(ASRM)to jointly learn adaptive similarity and predicted matrix by solving a similarity regularized optimization problem based on a new Augmented Lagrangian Multiplier(ALM)algorithm.Extensive experiments show that ASRM performs consistently well on a variety of real networks,and outperforms other state-of-the-art methods.Moreover,we shown the convergence performance of the ALM-based algorithm.In conclusion,ASRM is promising and useful for predicting missing links in complex networks.