CLASSIFICATION OF CONTEXT-DEPENDENT RECOMMENDER SYSTEMS FOR RECOMMENDING POINTS OF INTEREST

Zhdanov Pavel Sergeevich

To date, context-sensitive recommender systems for recommending points of interest are effective tools for analyzing user behavior: they not only reveal user preferences, but also allow you to generate relevant recommendations for locations that will be interesting for the user to visit under given conditions. Initially, for such recommendations, various modifications of models based on matrix factorization with the addition of contextual features were used, however, with the development and mass distribution of deep learning models, researchers began to actively use neural networks in context-sensitive recommender systems to recommend points of interest, which made it possible to deduce the quality of the algorithms. to a whole new level. This study offers a comprehensive review and analysis of context-sensitive recommender approaches for point of interest recommendations based on matrix factorization and deep learning to form a common understanding among researchers interested in the subject. The paper describes various modifications used in the creation of matrix factorization models and changes in neural network architectures to work with various context factors. Also in the study, in addition to the classification of models, the features of each class of models, their advantages and disadvantages are presented.

Keywords:context-sensitive recommender systems; recommendation of points of interest; matrix factorization; neural networks