A Genetic-Algorithms Matrix-Factorization-Based Recommender System Model Using Tags

Recommender systems (RSs) are useful technology that can alleviate the problem of overload of information provided to users. In this research, we build a new RS, and we name it ETagMF. ETagMF is an Evolutionary-based Tags-based Matrix Factorization (MF) model. Recently, researchers have introduced several MF-based approaches to improve the performance of the recommendation. MF is a multiplication of the items and user preference matrices in order to predict the unknown rating of items. ETagMF replaces the latent factors with the tags. It then uses Genetic Algorithms to predict the unknown rated items. It aims at improving accuracy, speeding up the recommendation process, increasing transparency and interaction. As far as we know, we have not found in the literature any similar work that applies the evolutionary algorithm and the tag-based MF techniques to predict values for the unknown items. Experimentally, we use Movielens dataset, and we show that ETagMF achieves good results. Furthermore, it outperforms the other competitive and similar state-of-the-art evolutionary-based collaborative filtering RSs. We compare ETagMF versus the Evolutionary Based Matrix Factorization (EMF) method done by Navgran using 500 latent factors for EMF and 500 tags for ETagMF. EMF is similar to ETagMF except that ETagMF uses tags and EMF uses latent factors. We show that ETagMF outperforms EMF, and it is 2.41 times faster than EMF and 12 times faster than the traditional MF method.