Exploring the applications of user-expertise assessment for intelligent interfaces

An adaptive user interface relies, to a large extent, upon an adequate user model (e.g., a representation of user-expertise). However, building a user model may be a tedious and time consuming task that will render such an interface unattractive to developers. We thus need an effective means of inferring the user model at low cost. In this paper, we describe a technique for automatically inferring a fine-grain model of a user's knowledge state based on a small number of observations. With this approach, the domain of knowledge to be evaluated is represented as a network of nodes (knowledge units—KU) and links (implications) induced from empirical user profiles. The user knowledge state is specified as a set of weights attached to the knowledge units that indicate the likelihood of mastery. These weights are updated every time a knowledge unit is reassigned a new weight (e.g., by a question-and-answer process). The updating scheme is based on the Dempster-Shafer algorithm. A User Knowledge Assessment Tool (UKAT) that employs this technique has been implemented. By way of simulations we explore an entropy-based method of choosing questions, and compare the results with a random sampling method. The experimental results show that the proposed knowledge assessment and questioning methods are useful and efficient in inferring detailed models of user-expertise, but the entropy-based method can induce a bias in some circumstances.