Математична база для створення гібридної нейронної мережі для рекомендацій з глибоким навчанням

Abstract

This work is devoted to the description of the mathematical framework that is needed to create a hybrid neural network for recommendations with deep learning. This approach to recommendations combines collaborative filtering and content processing of those elements on which training will be performed. The paper considers the mathematical side of the model which uses a loss function for training, a stochastic gradient descent (SGD) method and back propagation to adapt the model to minimize the loss. All these aspects described by mathematical formulas allow to write and train a neural network for recommendations, which can accommodate many parameters for recommendations and can be written in any platform and programming language