Research On Link Prediction Of Signed Complex Networks

Signed network has important research significance and application value in the field of machine learning and data mining research.Link prediction in signed network has attracted more and more attention of the researchers in many areas.Link prediction is an essential research area in network analysis.In recent years,link prediction in signed networks has drawn much attention of the researchers.To predict the potential positive and negative links,we should predict not only the existence of the link between the nodes,but also the sign and the probability of the existence of the link.In addition,the link prediction result should satisfy the balance and social status theory as much as possible.Most of the existing link prediction methods apply traditional unsigned link prediction methods on signed network.In those methods only positive links are concerned and negative links are ignored or treated as positive links.However,negative links may always play a vital role in analyzing the structure of the signed networks and the propagation of information in the networks.Therefore,it is necessary to design efficient methods specifically for link prediction in signed networks.To solve the problems mentioned above,we manage to find effective methods for link prediction of high quality in signed networks.The main work and research results are as follows:(1)An algorithm for link sign prediction based on Katz Index is proposed.Based on classic social balance theory,the algorithm applies Katz Index,which is a similarity index used in network link prediction,and takes topological structure features into consideration to make sign prediction in signed networks.Our experimental results show that this method greatly improves accuracy of the predicting results and saves computational time and cost.(2)Aiming at predicting both the sign of a link and the probability of the link with the sign,an algorithm link prediction on signed networks based on latent space projection is proposed.Taking the balance and social status theory into consideration,we define a balance/status weight matrix to reflect the balance/status constrains on the sign of the unknown links.We propose a model which combines the latent space and the balance/status constrains.An alternative iteration algorithm is proposed to optimize the model.The convergence and correctness of the iterative algorithm have been proved.Empirical results on real world signed networks demonstrate that the proposed algorithm can obtain higher quality predicting results than other algorithms.(3)A link prediction algorithm based on precision optimization is proposed.In this method,precision is treated as the objective function,and link prediction is transformed as an optimization problem.A group of topological features are defined for each ordered pair of nodes.Using those features as the attributes of the node pairs,link prediction can be treated as a binary classification where class label of each node pair is determined by whether there exists a signed link between the node pair.Then the binary classification problem can be solved by optimization on the precision.Empirical results show that our algorithm can achieve higher quality results of prediction than other algorithms.