The development of social networking services has aroused interest in predicting hidden information from a large amount of freely available public content. To predict user attributes, as a rule, information is used that the user left about himself in his profile. In this paper, we study and apply supervised machine learning methods with the teacher to the tasks of determining the age and gender of the user, using information from the user profile, as well as a method that extracts information from the social graph of the user and presents this information in the form of vectors - embeddings (DeepWalk). In addition, a graph neural network is implemented, which solves the problems of the DeepWalk algorithm. According to the results of the analysis, the use of graph embeddings gave an increase in quality both in the task of determining age and in the task of determining sex. The use of the graph neural network improved the quality only in the task of determining the sex of the user.