Abstract: In order to improve the accuracy of personalized movie recommendation, this paper proposes a hybrid recommendation algorithm, termed as CAMF-CM, which combines a decision tree model with a matrix decomposition algorithm containing user context information. By means of the matrix decomposition algorithm that incorporates context preferences, we shall obtain the initial movie recommendation list TOP-N. And then, a decision tree algorithm is used to perform the feature label training on the context data set LDOS-COMODA to obtain the user's movie preferences in a given context. According to the obtained TOP-N recommendation results, the user's selection tendency in a given context is collected via the decision tree model, and the TOP-N list is filtered again to obtain the final TOP-N recommendation list. A ten-fold cross-validation method is utilized to verify the proposed CAMF-CM algorithm, in which four algorithms are compared, including the MAE mean of the collaborative filtering algorithm, the basic matrix decomposition algorithm, the Baseline prediction algorithm, and the CAMF-CM hybrid algorithm. It is shown from the size of the MAE mean that the proposed algorithm can deal with the lack problem of interpretability of the results obtained by the traditional matrix factorization algorithms, and also overcomes the shortcoming that the traditional recommendation algorithm does not consider the situation. By the comparative selection of decision tree models in the context data set LDOS-COMODA, it is verified that the proposed CAMF-CM recommendation algorithm has higher accuracy than other algorithms, including the user-based collaborative filtering algorithms, basic matrix factorization algorithms, and Baseline recommendation algorithms.