Research On Dynamic Link Prediction Methods Of Social Networks

With the popularity of social network apps,the relationship between people in the real world can be abstracted as a link between users in a social network.The goal of link prediction is to predict whether new links will be formed between users based on the existing relationships of users and their attributes.Therefore,predicting new links between nodes in social networks can be considered as predicting new relationship between people.This research is of great significance.This thesis uses supervised methods for link prediction in social networks.Supervised link prediction methods use proximity between nodes to form the features for classifying links.However,existing supervised link prediction methods only illustrate the metrics that evaluate the links and do not offer the reason for selecting these metrics.Therefore,this thesis proposes a method to evaluate the influence of metrics between nodes on prediction result and use this as a basis for metric selection.Besides,the existing link prediction methods only consider the topology metric between node pairs to compose the feature of the link,but ignore the node attribute information contained in social networks.Therefore,this thesis proposes a link prediction method that considers both topology proximities and the attribute similarities between nodes based on the basic time series and stream link prediction methods.Finally,existing works only compares the accuracy of methods,and the results of link prediction can be displayed in a visible way.The Python is used to build a visualized simulation system.The proposed method is experimented and tested based on co-author dataset and microblog dataset.The result shows that the proposed method that considers attributes of nodes has higher accuracy than the method without such consideration and the visualized system can recommend the possible friend to a given user.