Abstract: The growth of a number of electronic devices and their capacities trigger the development of recommendation systems in general and adaptive interfaces in particular. Such systems collect information about user experience and analyze obtained data using the methods of machine-aided learning. Not ony the number of devices is growing, so is the number of users who often have different needs and demands. The author of the article pays special attention to the interface of USSD services. The purpose of the research is to develop an approach alllowing to decrease the average number of demands made by users while working with the text menu. To find a relevant solution, the author applied the methods of systems analysis and discovered a strong dependence between needs of each user and his/her context. Using the information about the execution context allowed to improve the existing methods and obtain better results. The author's approach is based on the two-stage clusterization: interface clusterization at the first stage and context clusterization at the second stage. To carry out testing the author used data collected during interaction with the users of USSD-services.