Optimising latent features using artificial immune system in collaborative filtering for recommender system

In collaborative filtering, the stochastic gradient descent (SGD) method is used to determine the latent features used in producing a non-negative N x M matrix of user-item ratings. The method is commonly used because it is straightforward in implementation and has a relatively fast running time. In this paper, we propose an artificial immune system approach to matrix factorization (AISMF) to optimise the latent features during the learning process. Artificial immune systems have the advantage of being dynamic, adaptive and able to learn an antigen in a few cycles. Therefore, they are well suited for the collaborative filtering of recommender systems. The performance of the AISMF is compared to that of the user-based and item-based neighbourhood clustering methods, SGD, Slope-one and Tendency-based methods. The results show that the AISMF converges faster to local minima for small to medium sized datasets and the AISMF ensemble performs better and faster, on average, on large datasets. The results also show that the AISMF ensemble is comparable to that of the SGD, user-based, item-based, Slope-one and Tendency-based methods in CF and can be used as an alternative learning and recommendation method in CF. (C) 2018 Elsevier B.V. All rights reserved.