Роль етапу оцінювання в каскадній системі рекомендацій

Abstract

The paper examines in detail the evaluation stage, also known as ranking, within the framework of the cascade architecture of recommender systems. An approach based on the use of deep neural networks for accurate prediction of ratings is proposed and analyzed. The architecture of the ranking model is investigated, including the use of multiattribute data to increase accuracy. Different configurations of the neural network and their impact on the final result are analyzed. It is shown that the use of a neural network approach at the evaluation stage allows significantly improving the quality of final recommendations by taking into account complex nonlinear dependencies between user and item characteristics..