Cognitive fuzzy number (CFN) considering the joint part of membership and non-membership degrees, is an interpretable information representation model to deal with the membership degree plus non-membership degree is larger than one. However, the CFN has two points for improvement: 1) the objective determination of joint degree; 2) the missing of hesitancy degree in the distance measure of CFNs. To tackle these challenges, we first add the hesitancy degree to the distance measure of CFNs. Based on the improved distance, the score function is proposed. Then, we set up a score-based consensus maximization programming model to determine the joint degrees of CFNs. Evidential reasoning (ER) approach is a method to deal with uncertainty based on a collectively exhaustive and mutually exclusive discernment frame. We introduce the ER approach to aggregate the cognitive fuzzy decision matrices of experts into a collective one, and then apply the ER approach to get the comprehensive evaluation value of each alternative. Alternatives are then ranked according to their comprehensive scores. We verify the applicability of the proposed CF-ER approach by an illustrative example. Sensitive analysis and comparative analysis are given to demonstrate the reliability of the proposed method.