Search

The work is related to solving the problem of personalizing Internet resources in order to increase the efficiency of organization and maintenance of psychologically safe user behavior in the Internet space. This problem is of particular relevance for the field of online education in the context of the continuous growth of the number of educational Internet resources and services, the development of content, which has a victimic effect on the moral and ethical aspects of the student's personality. Among the most promising approaches to its solution, currently stands out an approach that is based on predicting user preferences for generating content that meets their interests and expectations. In order to efficiently extract and process data, as well as to identify patterns that allow conclusions (predictions) to be made about specific user preferences, data mining and machine learning methods are used. The article proposes a method for predicting the preferences of the subject of training in intelligent assistant systems for generating recommendations and personalizing educational content and services. The principal difference of the proposed method is the transition from personal recommendations to group based on methods of collaborative filtering of users with similar preferences. To formulate recommendations, a preference evaluation model has been developed, based on the combined use of Item-Item CF and User-User CF methods, which will solve the cold start problem and improve the quality of recommendations for users with similar interests and behavior characteristics. Evaluation of forecasting is associated with a large dimension of the parameters of the learning model. To reduce the dimension of the search space preferences from a large amount of implicit data about user activity and improve the accuracy of the forecast, a bioinspired fish school search algorithm has been developed that is distinguished by scalability and the ability to process sparse data. Experimental studies of the efficiency of the algorithm were carried out on MovieLens test data and confirmed the high generalizing ability and accuracy of forecasting preferences.