Similarity Index-based Learning Link Prediction In Complex Networks

Link prediction has always been a hot issue in the field of complex network science.Most of the existing link prediction algorithms of complex networks are mainly based on a single similarity index.However,there has its own limitations and separation in their applications.Aiming at this problem,in this paper,under the framework of several algorithms of machine learning,based on integrating four similarity indicators,we investigate the link prediction problem of complex networks.The main research work includes the following two aspects.(1)Based on four similarity indexes(i.e.CN,LHN-II,COS+,MFI),we constructs feature vectors by extracting the characteristics of any two nodes in the complex networks.The random forest algorithm is applied to train the model,and a new link prediction algorithm for complex networks is proposed.Taking the American aviation network as an example and comparing the AUC values obtained by the previous researches with a single index,it is found that the AUC value and the stability for the algorithm provided in this paper are improved.(2)In order to further improve the accuracy of link prediction,the idea of ensemble learning is introduced into the link prediction of complex networks.According to the logistic regression algorithm and Xgboost algorithm,we construct two base models for the four indexes.Each base model achieves the probability feature of whether a node pair of the network has a connection or not.On this basis,using the logistic regression algorithm again,with the probability feature,the Stacking integrated model is trained.Finally,taking the real American aviation network and nematode neural network as two examples,compared with the existing algorithms,the experiments show that the AUC value of the proposed method has increased.And the method has better stability and recall rate.