An Improving Framework for Movie Recommendations with a Fusion of Deep Learning and K-Nearest Neighbor Algorithms

This paper investigates the application of two improved deep learning and K-nearest neighbor algorithms, respectively named as AutoDeepKNN and DeepKNR, in movie recommendation systems. We conduct experiments on two large-scale publicly available movie recommendation datasets, Movie-lens and Netflix. Detailed settings for experimental parameters, including data size, number of users, number of items, and the ratio of training and testing sets, were established. By utilizing various evaluation metrics such as precision, recall, F1-score, NDCG, and coverage, and comparing with baseline methods and traditional statistical models, the experimental results demonstrate significant performance improvements in all metrics from our proposed AutoDeepKNN and DeepKNR. Particularly, the AutoDeepKNN method, during the evaluation of precision and loss value, surpassed other methods significantly for 80% of the time.Our research not only enhances the accuracy of movie recommendations but also provides new methods and insights for other types of recommendation tasks.