<p>Deep non-negative matrix factorization (DNMF) is popular in multi-view clustering due to its hierarchical data-mining capacity. To improve clustering, numerous algorithms with diverse regularization terms or constraints have been put forward, yet existing methods have problems such as the inability to clearly distinguish shared and unique information, incomplete preservation of the global structure in the manifold space, and ignoring solution uniqueness. This paper presents the Comprehensive deep contrastive manifold regularized non-negative matrix factorization (CDCGONMF) for multi-view clustering. By using non-negative matrix factorization (NMF) to construct a deep framework, it explicitly separates shared and individual representations to condense shared information and eliminate redundancy. A novel contrastive manifold graph regularization is devised to maintain the global geometry of multi-view data, strengthening both similarity and dissimilarity relationships in the original high-dimensional space. Moreover, a quasi-orthogonal constraint is added to enhance solution uniqueness and sparsity, and an effective multiplicative update process supports the objective function. Evaluations on real-world datasets comparing CDCGONMF with benchmark methods demonstrate that CDCGONMF remarkably outperforms the state-of-the-art multi-view clustering methods.</p>