Developement of a Recommendation System and Study of its Accuracy Depending on Used Formal Methods

The rapid growth of data availability makes it difficult for users to effectively search for and filter the necessary information. As a solution to this problem, the technology of recommendation systems has been actively developing in recent years and offers users personalized and relevant recommendations. Over the past few years, most of the new recommendation models invented by researchers and employees of large IT campaigns are based on deep learning. Such models have become very popular as they allow you to capture more complex, non-linear interactions between the user and the element, which in turn leads to increased accuracy of recommendations. The purpose of this work is to implement recommendation models and compare their effectiveness and accuracy using the publicly available unified MovieLens dataset. Special attention in this work is paid to the evaluation and comparison of modern recommendation models based on deep learning. As a result of the work, a traditional model was implemented, along with several different models based on deep learning using the Python programming language. Experiments were also conducted, as a result of which estimates of these models were obtained using different metrics. Then a comparison of the models was carried out. As a result, the most effective model was identified among the considered ones. Model testing was also carried out, which consists of providing recommendations to the test user. The results obtained in the work and the software implementation of models and experiments can be used in the future when conducting research on the accuracy of new models and comparing them with earlier models. The volume of work is 46 pages. The paper contains 21 figures and 8 tables.