Evaluation of ERST - An external representation selection tutor

This paper describes the evaluation of ERST, an adaptive system which is designed to improve its users' external representation (ER) selection accuracy on a range of database query tasks. The design of the system was informed by the results of experimental studies. Those studies examined the interactions between the participants' background knowledge-of-external representations, their preferences for selecting particular information display forms, and their performance across a range of tasks involving database queries. The paper describes how ERST's adaptation is based on predicting users' ER-to-task matching skills and performance at reasoning with ERs, via a Bayesian user model. The model drives ERST's adaptive interventions in two ways - by 1. hinting to the user that particular representations be used, and/or 2. by removing from the user the opportunity to select display forms which have been associated with prior poor performance for that user. The results show that ERST does improve an individual's ER reasoning performance. The system is able to successfully predict users' ER-to-task matching skills and their ER reasoning performance via its Bayesian user model.