Research On Discovery And Analysis Of User Online Social Circle On Social Media

Social media becomes an important platform of information dissemination,and users can follow many information sources and interact with their friends.So that makes large user relationship networks appearc on social media.Also,there are many users' real life friends on these social networks.In traditional social science,it is difficult to research on individuals' relationships since the limitation of survey and data sparsity.The data can be accessed easily on social media,so that makes it becomes a new platform of research on human strong ties and social circles.The members in a social circle are homogeneous and friends in social circles can impact individuals' behavior each other.Therefore,research on user social circles is an important foundation of user behavior analysis.Every user has many friends on social media and these friends have different tiestrength,but these different tie-strength is difficult to be distinguished.And every user also has many social circles on social media,such as classmates in high school and classmates in university.A users' social circles are typical forms of the user's strong social ties,these social circles can represent users' different social dimensions.However,social circle is a subjective and private issue,so every user can just be familiar with their own social circles.So researchers are difficult to collect structures and social meanings of a user's social circles.The aim of this dissertation is to solve relevant problems about social circle discovering and analysis.Firstly,we focus on user online social circles based on user relationships.Every social circle is a category of user strong ties,so connections within a social circle should be very dense.Based on this principle,the dissertation propose a social circle discovery algorithm based on agglomerative clustering.The algorithm computes users' similarity based on users' social relationships.The model can discovery users' different social circles precisely.For the problem of data sparsity,we build a crowdsourcing platform for user annotating their own social circles.It has been collected enough real data for relevant research on user social circles.Secondly,we discuss about social circle discovering via multiple dimensional features.The members in a same social circle not only have dense connections,they but also have homogeneous profiles.Now social circles detection algorithms based on both networks and profiles can get effective performance.However,these methods can not combine two kinds of features naturally.This dissertation proposes a joint model of latent factor based on matrix factorization,and the model can learn user vectors with combining user different dimensional features.Comparing with models by user single feature,this model can discovery users' social circles more effectively.Thirdly,the dissertation gives tags to users' social circles.As a category of social strong ties,every social circle has its social meanings.Every member of the social circle has her/his own tags,and some members' common tags can represent the social meanings of a social circle.But a lot of users just have few tags and some users have even no tags,this problem leads it is difficult to detect tags of social circles.The dissertation proposes an algorithm of social circle tags detection via multiple linear regression.The model considers both features of users' tags and network topology.It gives a weight for every tag in a social circle,tags with larger weights are more likely be tags of social circles.Comparing with baselines,this model avoids data sparsity and improves the performance of social circle tags detection.Finally,we focus on user profile completion via users' social circles.User profiles are users' significant features on social media.User profile completion is a popular area which solving the sparsity of user profiles.Most of existed profile completion methods are based on the feature of text,the diversity of text leads a lot of noise for completion results.Social circles are users' strong ties and different social circles can represent users' different social dimensions.So the dissertation proposes a user profile completion method via non-negative matrix factorization.The model completes users' different social dimensional profiles and guarantees the diversity of user profiles.And the model also achieves better performance than related baselines.In summary,this dissertation describes research on some techniques of user social circle discovering and analyzing.And these techniques can be apply to user social circle analysis on general social medias.The research of the dissertation has achieved some preliminary results,which we hope can be helpful to research on user analysis.