Mining user interests on microblog is the basis for personalized recommendation and community classification. A descriptive model of microblog network is proposed based on the in-depth analysis over the characteristics of microblog in the work, revealing properties of multi-mode microblog. The representation and mining method of profile-based static user interests and microblog post-based dynamic user interests are proposed respectively according to the characteristics of microblog network. For mining inactive users with little profile and few microblog posts, a method of follower-based interest mining is proposed. In the case study of Sina microblog, users in fashion, business management, education, military and culture are selected for experimental analysis and comparison of interest mining and similarity calculation. Experimental results show that the proposed representation and mining method can effectively improve user interest mining comparing with other state-of-the-art methods.