A hybrid multi-criteria recommendation algorithm based on autoencoders.

Multi-criteria recommender systems provide efficient solutions to deal with information overload problem by producing personalized recommendations considering multiple criteria. Even though multicriteria recommender systems provide more accurate and personalized recommendations to their users compared with traditional recommender systems, sparsity becomes a major problem for such systems due to the increasing number of criteria. Due to the lack of corated items among users, finding out neighbors and producing accurate predictions become harder. Especially similarity-based multi-criteria recommendation approaches are significantly affected by the sparsity problem. Thus, aiming to minimize the negative impacts of that problem, a hybrid similarity-based multi-criteria recommendation method, that utilizes complex, low-dimensional and latent features obtained from both reviews and criteria ratings by autoencoders, is proposed in this work. The empirical results performed on a real data set with a sparsity percentage of 99.7235% show that the proposed work can provide more accurate predictions compared with other neighborhood-based multi-criteria approaches. [ABSTRACT FROM AUTHOR]