Multi-view data commonly existed in different practical applications, such as disease diagnosis, multimedia analysis and recommendation systems. However, sometimes the collected data may have samples with missing information in some views. Incomplete multi-view clustering (IMVC) is an important unsupervised analysis method for this kind of data. However, most of the existing IMVC methods focus on the consistency of different views while ignoring the potential inconsistencies between them. To solve this problem, we propose a novel graph learning based multi-view clustering method by taking into account both of consistent similarity and inconsistent similarity between samples under different views. Subsequently, a missing information completion method based on multi-view consistency is proposed for incomplete data analysis. Extensive experiments show the promising performance of the proposed method in comparison with several state-of-the-art algorithms.