Research On Evolution Of Social Networks Based On Link Prediction

The rapid development of social networking services has profoundly affected people's daily lives.The massive users in the social platform constitute a real-world network mapping,but there is still a lack of systematic understanding of many phenomena and evolution laws of this network.Studying the evolution of social networks can help to understand the network generation mechanism and predict network connection trends.In the context of social network,this paper studies the evolution of social networks from the perspective of link prediction.The main contributions of this paper are as follows:(1)From the three aspects of prediction accuracy,time consumption and stability,the commonly used link prediction methods are fully compared and analyzed.There are many methods currently used for link prediction,but the comparative analysis of these methods is lacking in current research.Due to the complexity and diversity of the network structure in reality,it is impossible to apply any method to all networks.Therefore,by comparing and analyzing the existing methods,it is helpful to judge the applicability of these methods.In this paper,the accuracy and time consumption of some common link prediction methods are compared on three generation networks and five empirical networks,and the stability of these link prediction methods is analyzed from the perspective of link noise and link loss.It is found that the network structure has a significant influence on the accuracy of prediction.The link prediction method based on the global structure has a large time consumption,but has better stability in the environment of link noise and link loss than the local link prediction method and the quasi-local link prediction method.(2)The link connection characteristics of static networks and dynamic networks are analyzed respectively.Each link prediction method is obtained based on the connection characteristics of links in the network.The link prediction method that can describe the link connection characteristics more reasonably and effectively can undoubtedly improve the accuracy of link prediction.Therefore,this paper analyzes the link connection characteristics of static networks and dynamic networks.The analysis finds that the structural characteristics of the nodes at both ends of the link have different characteristics depending on the network.Therefore,only considering the indicators of certain structural characteristics can not achieve good results.Generally,the closer the two nodes are,the easier it is to generate link between them,and the nodes connected to the internal links and external links in the dynamic network have obvious laws on the closeness centrality and betweenness centrality.(3)Based on the degree of node and spatial distance,an evolution model of spatial interaction network is proposed.Geospatial as a bridge between online and offline,affecting the interaction of users in social networks.This paper proposes a method for constructing spatial interaction network evolution model based on the preference attachment link prediction method for the urban interaction network formed by the information dissemination process on social networks.By comparing and analyzing the real city interaction network extracted from the We Chat webpage information dissemination data,it is found that the evolution network can match the real network well,reflecting that the model can not only capture the attributes of the real urban interactive network,but also reflect the actual characteristics of the interaction between cities.(4)The spatial interaction network model is expanded from the economic and demographic perspective.In a spatial interaction network,a node represents a specific area of the space,has a specific realistic meaning,and also contains rich attribute information.In general,important urban nodes are usually the political and economic centers of a region,so the GDP and population of a city can be used as a measure of the importance of the city.Through analysis,it is found that population,GDP,network degree and betweenness centrality are significantly correlated.Therefore,a method for constructing spatial interaction network evolution model based on population and GDP is proposed.By comparing the evolution network with the real network,the evolutionary network is almost identical to the actual network in terms of basic topological features.In the community structure,it can be found that the real network has a high similarity with the evolutionary network generated by the proposed model.In summary,this paper analyzes the common methods of link prediction,analyzes the connection characteristics of network links,and fully exploits the link characteristics to provide empirical support for constructing evolution models.Then,based on the spatial interaction network case formed by the information dissemination process on the social network,an evolution model for the spatial interaction network is constructed.The research results of this paper have important theoretical and practical value for information management and information recommendation on social networks.