Abstract: In order to improve the effect of multi-view clustering, a deep matrix decomposition algorithm in multi-view clustering is studied. A new low-rank deep matrix decomposition model in multi-view clustering is established by using low-rank constraints to capture potential low-rank data structures while removing redundancy and noise, and by introducing graph regularization to preserve the geometric structure of data in each view. The model is solved by the alternate direction multiplier method and the convergence theorem is given. Simulation results show that the algorithm is feasible and effective.