Target-Based Versus Customer-Based Genetic Algorithms and Matrix Factorization Recommender Systems

Recommender systems (RS) are a practical choice to help the customers choose the targets (i.e. book, restaurant, movie, TV, mobile phone, etc) they like in a short time. Collaborative filtering (CF) and matrix factorization (MF) techniques are common methods that are used in the RSs. The CF technique considers the rates of the customers to the targets. It recommends targets I for customer x if I have been rated high by similar customers to x. The MF technique splits the rating matrix into two matrices, customer-factor and factor-target, such that their targets nearly obtains the rating matrix. Navgaran RS (NRS) is a RS that uses the genetic algorithms (GA) to predict the uknown entries in the customer-factor and factor-target matrices in order to make recommendation for the active customer (i.e. the customer who wants recommendation). NLM is a GA-based CF RS that hybridizes both the neighbourhood and the latent factor models to predict targets for the active customers. NLM is a customer-based RS and improved NRS. In our project, we have modified NLM to make it target-based instead of being customer-based. We show that our new RS, NLM
u
, outperforms NLM and NRS in solving some instances.We experimentally used the datasets: MovieLens, FilmTrust, and CiaoDVD. We show that NLM
u
has better precision value than NRS in MovieLens. It is faster and has less error than NRS and NLM in FilmTrust, and it is faster and has less error than NRS in CiaoDVD.