Evolutionary extreme learning machine based collaborative filtering

Abstract

Research in the field of collaborative filtering (CF) has demonstrated its importance and effectiveness compared to other recommendation engines such as content-based and hybrid recommendation systems. However, there is ongoing research in the field of CF to improve the list of recommended items and generate accurate recommendations when dealing with sparse datasets. To enhance the prediction quality of recommender systems, a method based on an evolutionary extreme learning machine (EELM) for CF has been proposed to address the issue of sparsity. First, the dataset is pre-processed by filling in missing rating values in the rating database. Additionally, the dataset is trimmed by removing items with a limited number of ratings. This tailored rating database is then utilized with the extreme learning machine(ELM) technique to improve the quality of recommendations in the domain of CF. Furthermore, to enhance the accuracy of the proposed recommendation method, an evolutionary genetic algorithm(GA) was employed to train the parameters of the ELM-based model. The variants of the proposed scheme are compared to traditional recommendation methods by computing metrics such as mean absolute error (MAE), root mean square error (RMSE), precision, recall, and F-measure. Empirical analysis consistently indicates that the proposed approach outperforms the traditional CF-based recommendation method. The average accuracy metrics MAE and RMSE are reduced by factors of 9.69% and 32.83%, respectively, using the proposed technique. Additionally, the proposed scheme improves the classification accuracy of the recommender system by an average of 11.54%.