This paper explores the improvement of movie recommendations through neural networks, examining a Graph Neural Network (GNN) and a Deep Neural Network (DNN) with Softmax. The evaluation emphasizes how effective the GNN is at recommending films. The GNN skillfully records complex user-item interactions by utilizing graph structures and graph convolution to decipher user-movie relationships. Because of its capacity to extract latent information from the graph representation, it can provide more accurate and customized suggestions by providing a deeper understanding of user preferences and movie traits. The study presented in this paper aims to provide insights into the selection of the optimal model architecture for movie recommendation tasks by comparing the effectiveness and accuracy of both DNN and GNN based methods. The study shows that, although the DNN performs admirably with Softmax, it is less capable than the GNN in understanding the intricate networked relationships present in the movie dataset. This study also demonstrates how well the GNN models the complex interactions and provides better suggestions than conventional DNNs. The experimental results (using the performance metrics—precision, recall, F1-score and root mean square error (RMSE)) presented in this paper highlight that the ability of the GNN to decipher graph-based user-item associations signifies an effective way to improve the effectiveness of movie recommendation systems. The GNN model achieved lower error values and higher precision, recall, and F1-score compared to the DNN-based model. This study also revealed insights such as the automated design of the GNN model and fine-tuning of its hyperparameter as possible future research to strengthen its superiority.