Link Prediction In Complex Networks Based On Structural Properties

The problem of link prediction is an open issue in data mining and knowledge discovery,which attracts researchers from disparate scientific communities.The study of link prediction has a profound scientific meaning in understanding the organization and evolution of real-world networked systems.On the other hand,excellent link pre-dictors have a broad range of applications in a variety of domains,such as checking possible protein-protein interactions in biological networks,recommending promising candidate friendships for users in online social networks,and providing personalized recommendations in E-commerce systems.In this thesis,we intensively investigate the problem of link prediction from dif-ferent types of networks.To be more specific,the main contents and results of this thesis are summarized as follows:1.We reexamine the role of network structures in predict missing links from the perspective of information theory,and propose an information-theoretic model to incorporate multiple structural features.Based on the proposed model,we em-ploy a local structural characteristic of nodes,i.e.neighbor set,to develop a novel predictor named Neighbor Set Information(NSI).According to our experimen-tal results,the NSI index performs well in twelve real-world networks,compared with other typical proximity methods.Taken the NSI index as an example,an in-depth discussion on the information-theoretic model is given.2.We present a weighted mutual information model for weighted networks on the basis of mutual information from local network structures,which benefits from both structure and weights information.Empirical experiments are performed on four real-world weighted networks,and results indicate that the proposed model can provide more accurate predictions than not only traditional un-weighted methods but also typical weighted approaches.Furthermore,a different point of view on the effects of weak ties in link prediction is revealed.3.We develop a novel method for weight prediction based on the local network structure,namely,the set of neighbors of each node.The performance of pro-posed method is assessed in two cases.In the first case,some links and their weights are missing together,while in the second case all links are known and only weights of some links are unavailable.Empirical experiments on six real-world networks demonstrate that our method can provide accurate predictions of link weights in both cases.