The large number of movie additions, as well as the amount of user information collected, have raised the demand for improved and more effective recommendation systems. Algorithms such as traditional collaborative filtering or content-based filtering have been known to be effective; however, they suffer from the problems of sparsity of ratings and cold start. The main focus of this paper is to examine how deep neural networks can improve the accuracy of movie suggestions based on their interactions with collaborative filtering and content-based filters. Suggested a new structure for the model, which is a combination of a multi-layer perceptron for the user-item interaction and a convolutional neural network for movie meta features and users’ preferences. Tested the proposed model on a dataset containing 1 million ratings from 10,000 users across 5,000 movies. The results show that the proposed deep learning model outperforms the applied baseline methods, with a 15% improvement in prediction accuracy and a 20% increase in recommendation diversification. This indeed provides a solution to the cold-start problem by utilizing content features for the new items and users. Also, some interesting findings include the fact that dependencies between items and users—represented in the model as non-linear and latent—result in better recommendation accuracy. Based on these results, the use of deep neural networks can be considered a promising line of development for movie recommendation systems, which enables increasing user satisfaction and engagement in the applications for watching movies online or in a streaming manner.