It is rarely considered that the objective evidence source is disturbed by outside world and the evidence information is derived inefficiently in current research. To solve above problems, a method is established to gradually derive subjective evidence information from decision experts’ knowledge, experience and intuition based on dynamic knowledge matrices. After that, an inference theorem for transforming dynamic knowledge matrices to basic probability assignment functions is presented, and an interactive method for fusing subjective evidences is proposed to balance the contradictory relationships between inference costs and inference efficiencies by introducing a conflict weight threshold value. Simulation cases show the proposed method is efficient and applicable.