Social Network Classification System Based On Network Representation Learning

With the acceleration of global informatization,the Internet has changed people's way of life.Online social network has replaced the traditional way of interaction.More and more users tend to use social software for instant messaging,information sharing,hotspots discussion and other behaviors.However,each coin has two sides.Online social network is also a double-edged sword.It provides users with convenience while it's also quietly revealing users' privacy.The development of online social network has bred users' desire of showing themselves.Users are keen to share dynamic information in social software,no matter mood essays or travel updates.They are eager to communicate and interact with their social friends.Different dynamic information of users has its corresponding audiences.But data can't automatically identify the corresponding objects by itself.As a result,the dynamic information released by users is often relayed in different groups,which is out of the relative scope of privacy and leads to the leakage of privacy information.Therefore,the classification of social network can not only provide convenience for users in dynamic information sharing,but also limit the information sharing within the scope of users' intentions to effectively protect users' privacy.In this paper,we propose the automatic classification of social network from the perspective of privacy protection,and research the automatic classification scheme of social network for WeChat social platform.Starting from network representation learning and combining with natural language processing,this paper transforms the social network classification problem into the word vector classification problem.The main contributions of this paper are as follows:(1)We research the operational mode of online social network,users' behavior patterns and users' privacy protection needs.We propose the problem of automatic classification of social networks from the perspective of user privacy protection,as well as the issue of automatic publishing within a limited range according to the content of dynamic information.(2)We propose a model for quantifying the intimacy of online social network,and present a method for obtaining a list of relationship weights.To be specific,we analyzed the development status of online social network and users' interaction mode in social software,such as “thumb-up”,reposts and comments in MircoBlog,as well as “thumb-up” and comments in WeChat Moments.According to this,the intimacy relationship in social networks is divided into five categories.We count the interaction behaviors of friends with different intimacy relationships,and obtain the specific weight values corresponding to different degrees of intimacy by means of social survey.By taking the proportion of all interaction modes as the input of support vector machine and the weight value as the output,a high-precision intimacy classifier is trained for the classification of intimacy.Based on the collected interaction information in user's WeChat Moments,the classifier is used to classify the relationships between friends into corresponding category,so as to obtain the quantitative weight corresponding to the intimacy of all friends.Finally,we obtain the list of relationship weights.(3)Based on the list of relationship weights proposed,we present a high accuracy classification algorithm.Specifically,we used the second-order random walk that mentioned in Node2 vec algorithm to collect the information in the list,and then the node sequence set is obtained.The node sequence set is taken as the input of the natural language processing algorithm to get the word vector representation.In other words,we obtain the digital representation of the users' social friends.Furthermore,we use the elbow method to find the optimal clustering number of the word vector set.The word vector is clustered through the K-means clustering algorithm,so as to realize the automatic classification of social network friends.Finally,in the experimental result evaluation module,we obtain the visualization of the classification of social network on the two-dimensional plane by PCA dimensionality reduction.We define the classification accuracy to illustrate the classification effect,and propose the error processing method to update the classification result.We proved that the algorithm proposed in this paper can classify social friends with high accuracy,thus realizing the goal of classifying online social network with high-precision.