Research On Social Network Representation

As an important part in our daily life,social networks have gradually evolved from simple instant messaging tools to integrated application platforms such as breaking news recommendation,complex public services,and online payment,now it has become a virtual ecosystem encompassing various fields.Therefore,how to improve the user's stickiness to the social network platform and use the network resources efficiently have become the focus in both academia and industry.As the research basis of network community division,friend prediction,resource recommendation and many other network problems,social network representation has high universality and research value.This thesis will focus on how to represent network users accurately and efficiently.The content is as follows:1)The research related on network representation was studied widely,the characteristics of social networks were also analyzed,and then comes the introduction of word-embedding model word2 vec in natural language processing into social networks.The traversal network node forms a sequences so as to simulate the input sentence in the model training.In order to improve the traverse efficiency,a network node traversal rule with a constraint strategy was proposed,which reduces the traverse times according to node distribution and the number of generated sequences,along with improving the timeefficiency of the overall network representation.2)Inspired by the idea of tag propagation,this thesis proposes network representation algorithm of tag-based propagation and the key nodes of social networks based on the clustering characteristics of social network structures and generation of network sequences.That is,by finding key nodes with higher similarity,mining common neighbors of nodes and modifying the word2 vec model,making it possible fog nodes existing in the same community to carry similar “tags”.In this way,network representation could be improved though the method similar to tag propagation.3)The social network has structural characteristics and similar interactions such as likes and forwards.Therefore,A network representation model based on interaction behavior was established to improve the network representation effect.Finally,experiments were conducted using the BlogCatalog dataset and microblogging dataset,and the algorithm was compared with popular network representation models.The result shows that,by restricting network traversal times,the efficiency of network representation can be improved through network nodes` power-law distribution;The accuracy of network representation can be also improved by using network key nodes and users` interaction behavior.