User modeling in human-computer interaction tasks
Abstract
This research had two objectives. The first objective was to model a human-computer interaction task via a neural network model. The second objective was to use the model to identify the behavior of subjects as they performed the task and provide instructive feedback based upon that identification. Two experiments were performed. In the first experiment data from eight subjects were collected as they performed three computer debugging tasks while using a software debugger. The data were used to develop and train a neural network model for the debugging strategies. The network used the debugger options and computer program state information as inputs and mapped them to seven outputs. The outputs were the debugging goals of good and poor comprehension, good and poor code localization, and good, intermediate, and poor hypothesis-test activity. The network was able to learn 95% of the training data. In the second experiment, help messages were linked to the debugging outputs. The experiment evaluated the ability of the network to recognize the behavior of 24 new subjects and tested the effectiveness of the help. The network was able to recognize 80% of the test data; 86% of ambiguous patterns (where no identification could be made) were not considered. The errors made by the network were not severe. In the debugging task, experts outperformed novices, resulting in a significant difference in the time to locate bugs. Subjects who had help had more good comprehension, good localization, and intermediate hypothesis-test activity than those who did not. Subjects reported the help encouraged them to try options from the software debugger they were using that they might not have used otherwise. They also indicated the help caused them to rethink the way to go about debugging. This research has implications on methods to more accurately determine user behavior in human-computer interaction tasks. Once this behavior is recognized, feedback can be given which points out inefficiencies in completing the tasks so that performance improvements can be made.
Degree
Ph.D.
Advisors
Eberts, Purdue University.
Subject Area
Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.