User Profile Completion with Online Social Circles

User profile is a kind of most effective feature and cue of user behavioranalysis. However, most ofusers'profiles are incomplete and untruthful, and users' tag and profile data are rathersparse on social media. It makes profile completionbecomeapopularresearch topic. Users have manystrongrelationships online, and theserelationshipsformmany social circles, such as classmate, family and colleague. A user'sdifferent online social circlescanrepresentthisuser'svarious social attributes. We focus the task of user profile completion in thispaper. We propose an algorithm of user profile completion via users' social circles by non-negative matrix factorization. The methodcandetectauser's multi-dimension social characteristicsfromvarious social circles. Wealsotransfer the concept of social circle to academic networks. A papermay cite otherpaperswithseveral intentions. Some of the referencesmaybe relevant according to the researchproblems and otherreferencesmaybe relevant according to the methodologies. The references of different topics formdifferent "referencecircle". By an academicpaper'sdifferent "referencecircle", our model candetect the paper's keywords in differentviews. The experiment on Facebook LinkedIn and Microsoft AcademicSearch show ourmethod has improved performance than state-of-art methods.