Robustness Analysis Of Social Network Link Prediction Algorithms

Complex network is an abstract form of complex system.It abstracts the individuals and associations in the system as nodes and links in the network.In the application of complex networks,one of the most important is link prediction,which is to predict the association existing in the network but not found or the possible relationship in the future.In link prediction,the attributes of network nodes and network structure are usually used.However,in real life,there are often a lot of noise information in the network,which interferes with our prediction of unknown links.Therefore,the study of the impact of network noise on link prediction,that is,the robustness of network link prediction,has become our research topic.The purpose of this paper is to find out the robustness rule of link prediction algorithm in noisy networks by studying the robustness of network link prediction,and to propose a link prediction algorithm with not only high link prediction accuracy,but also high robustness.Firstly,this paper summarizes three kinds of common noises in social networks,proposes the robustness index for AUC,rating score and precision,and analyzes the indexes and robustness of different link prediction methods under different kinds of noises on multiple social networks.It is found that the index based on the local property of the network has higher link accuracy,and the index based on the global property of the network has higher robustness.Then,based on the scale-free feature of social network,we combine the information of network degree with data preprocessing and link prediction to improve the robustness of link prediction.Finally,this paper proposes an improved sdne method in the weighted directed network,which combines the local and global information of the network at the same time.When the algorithm is used in link prediction,it is found that the prediction accuracy and robustness of the network are improved.