SUMMARY
The paper reports a study into recommendation algorithms and determination of their advantages and disadvantages. The method for developing recommendations based on collaborative filtering such as Content-Based Filtering (CBF), Collaborative Filtering (CF), and hybrid methods of Machine Learning (ML) has been improved. The paper describes the design principles and functional requirements to a recommendation system in the form of a Web application for choosing the content required by user using movies as an example. The research has focused on solving issues related to cold start and scalability within the method of collaborative filtering. To effectively address these tasks, we have used hybrid training methods. A hybrid recommendation system (HRS) has been practically implemented for providing relevant content recommendations using movies as an example, taking into consideration the user's personal preferences based on the constructed hybrid method. We have improved an algorithm for developing content recommendations based on the collaborative filtering and Machine Learning for the combined filtration of similarity indicators among users or goods. The hybrid algorithm receives initial information in a different form, normalizes it, and generates relevant recommendations based on a combination of CF and CBF methods. Machine Learning is capable of defining those factors that influence the selection of relevant films, which improves development of recommendations specific to the user. To solve these tasks, a new improved method has been proposed, underlying which, in contrast to existing systems of recommendations, are the hybrid methods and Machine Learning. Machine Learning data for the designed HRS were borrowed from MovieLens. We have analyzed methods for developing recommendations to the user; existing recommendation systems have been reviewed. Our experimental results demonstrate that the operational indicators for the proposed HRS, based on the technology of CF+CBF+ML, outperform those for two individual models, CF and CBF, and such their combinations as CF CBF, CF+ML, and CBF+ML. We recommend using HRS to collect data on people's preferences in selecting goods and to providing relevant recommendations.