Optimisation of SVD++ Method based on Adam’s Algorithm for Small E-Commerce Platforms

This paper presents the SVD++ method for predicting ratings in recommender systems for small ecommerce platforms. It is shown that the SVD++ method is based on the matrix factorisation method and uses gradient descent to optimise the objective function. A comparison of optimisation algorithms for setting up the model in SVD++ is carried out, in particular, the following algorithms were considered: SGD, AdaGrad, RMSProp, and Adam. The results of the analysis showed that the Adam algorithm was the most effective. The optimisation of the SVD++ method using the Adam algorithm is studied on the example of a test sample from the GroupLens» resource.