Date of Degree

9-30-2015

Document Type

Dissertation

Degree Name

Ph.D.

Program

Computer Science

Advisor(s)

Matt Huenerfauth

Keywords

Adaptive User Interfaces; Eye Movements; Eye Tracking; Human-Computer Interaction

Abstract

While evaluating user task performance with eye tracking has been examined within the field of Human-Computer Interaction (HCI), previous research has generally utilized eye tracking to understand how users perform a task. This dissertation defines a new paradigm by which eye-tracking research can be used in HCI: to predict whether users will be successful at a task, using a pattern classification model trained on their eye-tracking data.

This dissertation describes the experimental framework and it demonstrates (through laboratory experiments and machine-learning modeling) the feasibility and limits of predicting user outcomes based solely on eye movement patterns. Utilizing an online learning scenario as a proof-of-concept application of this technique, several rounds of eye-tracking data collection studies were conducted in which participants viewed multiple windows of simultaneous information content: a video of an instructor, lecture slides, a transcript of the speech, etc. After viewing the lesson while being recorded with an eye tracker, students responded to a test in which their understanding of the information content was measured. A variety of machine-learning approaches and feature selection techniques were used to explore the relationship between the eye movements of students and their success on the final test, and classification models that outperformed a majority-label baseline were successfully trained.

To test the limits of this approach, additional studies were conducted with modified user-interfaces that relaxed some of the homogeneity of information content and visual presentation that were rigorously maintained in the initial experiment. In addition, a follow-up study was conducted in which the strict temporal segmentation of the experimental session was randomized, to measure the robustness of the modeling approach to such perturbations of the data-collection and analysis methodology.

The results of this study can assist HCI eye-tracking researchers in developing new techniques for evaluating systems, e.g., by predicting users' task performance based on eye-tracking data. Further, this dissertation lays the conceptual and methodological groundwork for the design of intelligent systems for predicting which users may be struggling with a task, based upon an automatic classification of users into groups of high- or low-performers based upon an examination of their eye movements alone.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.