Movie Recommendation Systems Using Improved Non-Negative Matrix Factorization

Currently, there exist effective strategies for providing practical recommendations crucial in many industries, this encompasses a range of online platforms, such as electronic commerce, social networking sites, and various web-based services, as a result of the big data explosion. There are numerous customized movie recommendation systems available today that use publicly accessible movie datasets (like MovieLens) and provide enhanced performance measures (such the Root-Mean-Square Error (RMSE)). The ability to scale and the practicality of usage feedback and verification are determined through real-world implementation and assessment, they are two important difficulties that movie recommendation systems currently struggle with. This research suggests a novel approach using non-negative matrix factorization that resolves these issues through similarity and matrix factorization. Finding two matrices, both consisting of non-negative values, and their resulting product when multiplied together closely resembles the original matrix is the goal of non-negative matrix factorization. Additionally, it places the latent elements under non-negative constraints. The suggested approach offers novel update procedures to discover the latent variables for rating prediction. In contrast to the majority of collaborative filtering techniques, the suggested technique can forecast all the unknowable ratings. Its computational complexity is quite low, and it is simple to implement. Empirical research using the MovieLens shows that the suggested approach is more forgiving of the sparsity and scalability issues and produces good results.