Research On Inference Method Of User Attribute Information

In recent years,with the rapid development of the Internet and mobile Internet,social media,represented by social platforms such as Weibo and Weixin is also developing rapidly.Social media which promotes communication among users and facilitates users' acess to information,is becoming an indispensable part of people's daily life.The behavior of users on social media brings a large amount of social network data,which includes the user's attribute information data.However,due to users' unwillingness to expose privacy information or other reasons,the user's attribute information is often missing or incomplete,which brings difficulties to the research of online social network.Therefore,some researchers began to study the inference of user attribute information in social networks,so as to supplement and restore the missing and incomplete parts of user attribute information.However,the current attribute inference methods mainly solve the attribute inference problem based on the unweighted network,without considering the strength of the relationship between user nodes,in attribute inference,neighbors with strong relationship should play a greater role than neighbors with weak relationship.In this paper,attribute inference is studied in the social network composed of Weibo users.Firstly,this paper introduces several traditional attribute inference algorithms,and investigates the effect of traditional attribute inference methods on experimental data sets.Secondly,this paper uses two methods to determine strong and weak relationships in social networks,one method is to set similarity threshold,the other method is based on community discovery algorithm.In the social network that determines the strong or weak relationship,we use the weighted voting method of the strong neighbor,to complete the inference task of attribute information,the strong neighbor refers to the neighbor node that has strong relationship with the unknown node and the neighbor node's attribute is known.In addition,the Node2 vec algorithm is used to train the vector representation of each node in the network,and the attribute inference problem is regarded as the node classification problem.The node representation vector is regarded as the feature of the node to train classifier,and the attribute inference problem is solved by the classification method.Furthermore,the feature of strong neighbor attributes is added to the feature of nodes,and the classifier is retrained and the node is reclassified.Experiments show that weighted voting based on strong relationship and the increase of attribute information characteristics of strong relational neighbors have a positive impact on attribute information inference,which verifies the rationality of the proposition that strong relational neighbors should play a greater role in attribute inference of unknown nodes.