Composing user models through logic analysis.

Abstract

The evaluation of tutorial strategies, interface designs, and courseware content is an area of active research in the medical education community. Many of the evaluation techniques that have been developed (e.g., program instrumentation), commonly produce data that are difficult to decipher or to interpret effectively. We have explored the use of decision tables to automatically simplify and categorize data for the composition of user models--descriptions of student's learning styles and preferences. An approach to user modeling that is based on decision tables has numerous advantages compared with traditional manual techniques or methods that rely on rule-based expert systems or neural networks. Decision tables provide a mechanism whereby overwhelming quantities of data can be condensed into an easily interpreted and manipulated form. Compared with conventional rule-based expert systems, decision tables are more amenable to modification. Unlike classification systems based on neural networks, the entries in decision tables are readily available for inspection and manipulation. Decision tables, descriptions of observations of behavior, also provide automatic checks for ambiguity in the tracking data.