Sparseness reduction in collaborative filtering using a nearest neighbour artificial immune system with genetic algorithms

摘要

In collaborative filtering, one of the main challenges that researchers face is sparseness in the data, which is caused by users rating fewer items as the number of items increase in the dataset. The effect is poor predictions and recommendations of items to users by expert and intelligent systems like the recommender system. This paper proposes a Nearest Neighbour Artificial Immune System with a Genetic Algorithm (NNAISGA) to perform fast data imputations to reduce sparseness. The main impact is sustained and reliable predictions as the number of missing data increases in the dataset. The second benefit is to help sustain or improve recommendations of items to users from common methods used in collaborative filtering such as the User-based, Item-based, Slope-one, Tendencies-based and Non-Negative Matrix Factorisation (NNMF) methods. We show that NNAISGA is the desired method for learning and imputation over the traditional genetic algorithm. The findings show that using the NNAISGA as a fast imputation method yields promising results. All of the methods, except for the User-based method, show significant improvements or sustained accuracies in terms of predictions and recommendations. (C) 2019 Elsevier Ltd. All rights reserved.