about users may be inaccurate or incomplete, which leads to distortion of recommendations. The received incorrect recommendations do not meet the interests of users, which reduces the credibility of the recommender system. Inaccuracies and distortions in user preference data result from shilling attacks [4]. Such attacks are used to change the sales of target items in the direction desired by the attacker. The essence of a shilling attack is to artificially change the ratings of the target items. As a result, a personal recommendation is formed with distortions and may not take into account the real preferences of users. The recommender system, using an incorrect personal list of goods and services, "forces" the user to choose those items that the attacker is interested in. In case of incomplete information about the user, recommendations are built for new or irregular users [5]. In such a cold start situation, there are no data obtained as a result of feedback from the user, since the latter has not yet made purchases or posted ratings. Therefore, at present, one of the trends in the development of recommender systems, which makes it possible to consider the incompleteness and inaccuracy of the initial data, is to