Decision theoretic perspective on optimizing intelligent help
The purpose of this study is to investigate a way to optimize the help messages provided by the intelligent agent from a decision theoretic perspective. The study proposed an expected time-based optimization rule utilizing the time associated with processing a help message and its time saving as the trade-off criterion of whether to present a help message or not. The interface for providing help messages was also investigated to see how users respond to different interface types of intelligent help messages according to user knowledge and task complexity. Six hypotheses were proposed regarding the effectiveness and user satisfaction of the proposed optimization rule and the help message interface, and then they were tested in the two experiments. In Experiment I, the expected time-based optimization rule was compared with no optimization rule and no help message. Thirty participants performed two tasks of troubleshooting a print quality issue using one of the three help provision conditions (optimized help messages, non-optimized help messages, and no help messages). In Experiment II, the effects of task complexity and user knowledge were tested on the two help message interfaces with forty subjects. The results of two experiments revealed the following: 1) the expected time-based optimization rule can significantly reduce user performance time and enhance user satisfaction for the complex task; 2) providing help messages for the simple task does not reduce performance time nor increase user satisfaction; 3) users are more satisfied with the help messages in a less intrusive interface regardless of task type and user knowledge. The combined results suggest that the intelligent help messages need to be presented in a less intrusive interface according to the expected time-based optimization rule to improve performance time and user satisfaction. The help optimization approach based on the expected time is expected to provide guidance as to where, when and why the intelligent help messages are likely to be effective or ineffective by utilizing quantitative predictions of value and cost of intelligent help messages in time.
Lehto, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our