Research On Online Social Network Link Prediction Methods

Link prediction is a hot research issue in many disciplines such as computer science and network science.The prediction of missing links or upcoming links between nodes helps to reconstruct the network structure and explore the laws of network evolution.At present,the idea of link prediction research is to use machine learning algorithm or network science theory to study the similarity between nodes,and then establish link prediction algorithms.However,most of the existing work based on different theoretical methods is relatively independent.Aiming at improving the link prediction accuracy in social networks,we combine machine learning algorithms and network science theory to study link prediction algorithms and systems.Aiming at the problem that the existing link prediction algorithm cannot fully utilize the network topology information,a link prediction algorithm based on Bayesian theory is proposed.First,we generalize the clustering coefficients of nodes and give the definition of clustering coefficients of paths.Then,we propose a new link prediction algorithm based on Bayesian theory by using Bayesian method to combine local topology information,global topology information and network evolution trends.Finally,we apply the algorithm to single-layer networks,multi-layer networks and time-varying networks.The experimental results show that the prediction accuracy of this algorithm is higher than several typical link prediction algorithms such as Common Neighbors algorithm,Katz algorithm and Local Na?ve Bayes algorithm etc.We propose a prediction algorithm that considers both the elements that facilitate link generation and the elements that suppress link generation.Through the statistical analysis of the social network data,we find that the communication demand between nodes is positively correlated with the probability of node connection,and when the communication capability(the number of paths)between nodes is sufficiently large,the probability of establishing new links is reduced.We improve the logistic regression algorithm and propose a link prediction algorithm by considering the communication demand and communication capability at the same time.The results of the experiment show that the proposed algorithm performs better than the typical logistic regression algorithm.In order to test the link prediction algorithm in real networks,a friend recommendation system is designed for social networks.The system uses the web crawler to crawl the QQ user information on Qzone.Then it employs the link prediction algorithm based on Bayesian theory to calculate the user relevance and makes friend recommendation based on the calculation.Experiments in real social networks demonstrate the application value of the link prediction algorithm based on Bayesian theory and provide a reference for further generalization and application of this algorithm.