Ph.D. THESIS DEFENSE: CoDeIn: A Knowledge-Based Framework for the Description and Evaluation of Reality-Based Interaction

May 7, 2007
5:00 pm - 7:00 pm
Halligan 127
Speaker: Georgios Christou, Tufts University
Host: Robert Jacob


Abstract: Today there is a plethora of new interaction styles that try to move the computer user away from the traditional computer setting, providing different modes and types of interaction that do not fit into the traditional keyboard-and-mouse-driven paradigm. These new interaction styles are not as connected as the previous generation, making comparison between interfaces designed in them difficult. Also, existing evaluation methods seem to encounter problems when dealing with these new interaction styles.

This thesis presents Cognitive Description and Evaluation of Interaction (CoDeIn), a framework that allows the description and evaluation of tasks designed in different interaction styles, and allows predictive performance evaluations and quantitative comparisons between these tasks. The evaluations performed with CoDeIn consistently outperform evaluations that are carried out using existing model-based evaluation methods. The evaluation process is based on the framework’s provision of structures that allow the quantification of terms like “good” or “bad” interfaces and for specifying users according to their task performance knowledge in some interface through the application of fuzzy logic techniques.

The argument is supported by two experiments that provide evidence towards the abovementioned claims. The first compares the performance of the framework with GOMSL on the evaluation of a task designed as a Direct Manipulation Interface and as a Tangible User Interface. This experiment also exemplifies how CoDeIn is capable of handling tasks that may include parallel actions on the part of the user.

The second experiment explicates how the framework can be used to predict the performance of different types of users (e.g. novices, experts, etc.), by using two interaction devices, one that was familiar to the participants and another that was not. The model built with CoDeIn closely predicts the participants’ performance with both interaction devices.