Automatic detection of users’ skill levels using high-frequency user interface events

Abstract

Computer users have different levels of system skills. Moreover, each user has different levels of skill across different applications and even in different portions of the same application. Additionally, users' skill levels change dynamically as users gain more experience in a user interface. In order to adapt user interfaces to the different needs of user groups with different levels of skills, automatic methods of skill detection are required. In this paper, we present our experiments and methods, which are used to build automatic skill classifiers for desktop applications. Machine learning algorithms were used to build statistical predictive models of skill. Attribute values were extracted from high frequency user interface events, such as mouse motions and menu interactions, and were used as inputs to our models. We have built both task-independent and task-dependent classifiers with promising results.