A novel scheme that employs content of posts and social relationships to analyze user opinion on microblog is proposed in this paper. Unlike traditional approaches focus on posts, this research regards user as an analysis unit to investigate sentiment classification on a specific topic. In aspect of textual sentiment classification, opinion of posts is classified with Bayesian or LibSVM tools. In addition, two types of social relationships (friends and fans) are adopted to construct an indirect graph of social network separately. The aim of this paper is to leverage user neighbors to overcome the challenge that posts are often too short and ambiguous to analyze opinions. Simultaneously, we deeply consider influential degree through interactions between two humans. In our experiment, Plurk, a popular microblog in Asia is employed as resource to achieve topic-dependent opinion analysis.