Multi-Batch Quasi-Newton Method with Artificial Neural Network for Movie Recommendation

Abstract

The recommendation methods are required in the e-commerce website to help users in the decision-making process. Social Network has a high influence on the movie-based opinions among the users and social network data can be applied in recommendation system to handle the cold start and data sparsity problem. Existing research has the limitation of the overfitting problem in the model and also has lower efficiency in cold start and data sparsity problem. In this study, a Multi-Batch Quasi-Newton (MBQN) and Limited Memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) gradient methods are implemented to solve these issues of the recommendation system. To solve the cold start and data sparsity issues, the MBQN method overlaps the consecutive sample. The L-BFGS gradient method is used for the parameter estimation in the machine learning technique and the MBQN method is used to select the subset of the training data. The different data points are used to evaluate the function and gradient at the start and end of the iteration. The proposed MBQN is evaluated using the artificial neural network (ANN). The nonnegative factorization method is applied to the data to reduce the overfitting issue in the ANN. The proposed MBQN method is evaluated using MovieLens datasets to analyze the performance of the recommendation system. The k-fold cross-validation is applied to evaluate the performance of the proposed MBQN with nonnegative matrix factorization. To analyze the robustness of the method, a large dataset is known as MovieLens 25 M dataset was applied to evaluate the performance. The experimental results show that the MBQN has a higher performance in the movie recommendation compared to the existing matrix factorization method. In addition to this, the results reveal that the MBQN method attains the ratio value of RMSE as 0.973 and the existing method attains the ratio value of RMSE as 2.776.