User Tag Recommendation In Social Networking

In recent years, Social Networking, such as FaceBook、Twitter, become moreimport, attracting more and more scholars engaged in research work related tosocial networking、user interest miningEtc. User tags can label user’s interests andconcerns. Identifying an individual user’s interests and concers can help potentialcommercial applicationsThis work start from user’s social network relations,by means of text analysistechnology to generate persionalized annotation tags that describe his intersts,concerns for studying Personalized Annotation Tags in social networking.This content of study covers three aspects. First, extracting candidate user tags,second, sorting candidate user tags by similiary between tag and user, diversity usertags recommendation.Extracting candidate user tags. When a user concerns another user as a friendbecause the have some common interest or identity attributes. We set personalizedannotation tags from concern friends as the initial user tags, To select candidate tagsby the importance of tag for covering user ’s interests. We conducted three ideas toselect tags, they are tfidf model, greedy methods, greedy set cover algorithm.Personalized Ranking candidate user tags. When we extract candidate tags, weignored that different concerned friends have different influence to user, andconcerned friends’s tweets represent his influence, we use tweets that user browsedto build user sematic model based on fact that one who has more influence to useroften writes good tweets.We combine similarity between user sematic model andtag with tag’s importance and make candidate tags personalized ranking.User tags recommendation diversity. Take into account that one often hasmany interests and candidate tags personalized ranking has a problem that tagsematic redundancy and can cover user’s interests. Based on a idea that if acandidate tag has some common sematic with user tags select, we should penalize it.We conduct MMR and MMRCluster experments to do user tags recommendation diversity.Experiments show that tags from concerned friends contain some tags that candescribe user’interests and identity; We remove much noise in candidate tags byusing tag’s importance of concerns friends to select candidate tags, User sematictags based on tweets user browsed can make user’s concerned friends different tosome extent. With the help of similarity between user and tag, we can obtain tagsthat describe user’s interests and identity more accuracily. Take tags diversity inaccount, we improve the problem that tags sematic redundancy and make tagsdescribe user more comprehensive.