Abstract: This paper analysed different topic models,and compared three extended topic models’ performance on mo-deling microblog user interestingness via three experiments．Experimental results show that TwitterLDA can apply to predict words on new unseen docuemnts and users,that the topics generated by AuthorLDA have a higher degree of differentiation,and that UserLDA and AuthorLDA can better reflect the users’ relationships in real social network．The work in this paper lays the foundation for further studying how the topic model is applied to the text mining applications of microblogs such as personalized recommendation,sentiment analysis and topic detection and tracking.

Key words: Topic model,User interest,Personalized service