Abstract It is indispensable for users to evaluate the trustworthiness of other users (referred to as advisors), to cope with possible misleading opinions provided by them. Advisors’ misleading opinions may be induced by their dishonesty, subjectivity difference with users, or both. Existing approaches do not well distinguish the two different causes. In this paper, we propose a novel probabilistic graphical trust model to separately consider these two factors, involving three types of latent variables: benevolence, integrity and competence of advisors, trust propensity of users, and subjectivity difference between users and advisors. Experimental results on real datasets demonstrate that our method advances state-of-the-art approaches to a large extent.