Research On Link Prediction And Community Detection Based On Node Affinity In Social Networks

Complex social activities constitute social networks,which contain flexible network structure and multiple roles,and are full of people's lives.In modern society,with the rapid development of social media and social ways,social network analysis becomes more and more important.Through social network analysis,the potential law and the evolution pattern of the network can be discovered to help us understand social phenomena.This paper starts with two key areas of social network analysis: link prediction and community detection.Both of them research the intrinsic evolution process of the network through the network topology structure,and explore the potential connections between network members.The goal of link prediction is to predict the links that will appear in the network,and community detection is to mine the sub-group structure with relatively close internal connections in the network.Considering the interaction between adjacent nodes has the most direct impact,this paper takes the network's affinity relationship as the starting point,and proposes a link prediction algorithm based on 1-hop nodes embedding.And in view of the opinion that users with high similarity tend to form closer connections,this paper proposes a coarsening networks framework for community detection.Network embedding has become a hot research topic in the field of network representation in recent years.However,most of the nodes embedding algorithms aim at the expression of node features,and ignore the customized design for link prediction.This paper aims at the task of link prediction,and focuses on distinguishing the affinity and sparseness of network relationships.Considering that the directly connected nodes in the network have a more direct and important influence on the upcoming links,this paper designs the 1-hop nodes sampling strategy.The feature expression of nodes is learned by the Skip-Gram model of word2 vec and is used to solve the link prediction problem.The results of the experiment show that the estimate accuracy of the link prediction algorithm based on 1-hop nodes embedding has greatly improved compared with the traditional link prediction algorithms based on similarity indices and the current mainstream nodes embedding algorithms.Especially for Vote dataset,the AUC value has increased by 11.6% on average.As an emerging research subject in the field of social computing,community detection has many related works.In this paper,we propose a community detection framework for coarsening network to minimize the negative impacts of anomalous links,which are generally neglected by existing algorithms.The homogeneity and cohesion of community structure determine that the links with low similarity are likely to exist between communities.At the same time,the node affinity has a potential impact on the similarity.Therefore,firstly,the links with low similarity are deleted to coarsen network.Then we detect communities based on the reduced network.Finally,experiments are carried out on the real dataset and artificial dataset,and comparisons between the algorithm presented in this paper and the mainstream algorithms are conducted.The results show that coarsening network can improve the accuracy of community detection while mining the fine-grained communities,and have more obvious advantages in dense graphs.