An Application of Differential Evolution Algorithm-based Restricted Boltzmann Machine to Recommendation Systems

Abstract

Due to growth of electronic commerce, currently, many customers prefer to buying products from the internet. Thus, recommendation system, like restricted Boltzmann machine (RBM), has become a good technique to recommend the right product to the potential customer. This can dramatically increase the customer loyalty. However, it is necessary to determine some parameters for RBM and enhance its computation performance. Therefore, this study intends to propose a hybrid algorithm which combines the cluster-based restricted Boltzmann machine (CRBM) with differential evolution (DE) algorithm to optimize the RBM’s parameters for collaborative filtering. The CRBM applies a clustering algorithm to determine the size and elements for each mini-batch gradient descent method for the RBM. The proposed DE-based CRBM algorithm is validated using four benchmark datasets. The results are compared with those of batch RBM, mini-batch RBM, clustering RBM, PSO-based and GA-based clustering RBM. The experimental results reveal that optimizing the RBM’s parameters using metaheuristic can obtain the better result. It also show that the proposed DE-based CRBM algorithm performs better than GA-based and PSO-based CRBM algorithms.