Automated customization of user interfaces

Author / Creator

Lerner, Barbara Staudt

Available as

Physical

Summary

Abstract: "With the prevalence of personal and small business computers, there is a proliferation of software houses developing tools for the general computing community. In such situations, it is ...

Abstract: "With the prevalence of personal and small business computers, there is a proliferation of software houses developing tools for the general computing community. In such situations, it is not possible for a user interface designer to design tools for a specific user community. This thesis addresses this problem by investigating mechanisms that allow user interfaces to be constructed so that they can customize themselves automatically to match the styles and habits of individual users. The proposed mechanisms use knowledge of the application domain, context in which the user is operating, as well as knowledge of the user's habits and idiosyncrasies to perform their customizations.

Customizations involve automation of routine tasks, interpretation of deviant input, and provision of active help. While domain and context knowledge are generally encoded directly in the underlying system, user knowledge is acquired by observing the user's interactions with the system. These interactions are analyzed to find patterns in the user's behavior which can be automated. In addition to automating the customization process itself, the thesis also investigates the use of automated evaluation mechanisms, called success/failure criteria. When an action is automatically performed, success/failure criteria are used to monitor subsequent user actions to determine if the automated action was acceptable to the user, or if the user undid the action.

The results of this evaluation are used as feedback to the customization process. In this way, the complete customization process can be done without direct user intervention. This includes identifying situations where customization is appropriate, determining what action to take, and deciding if the action was correct. The mechanisms proposed in the thesis were implemented in a Gandalf programming environment. The major results of experimentation are that automated customization resulted in a 7% decrease in the number of commands required to complete a task, and up to 25% reduction in the number of errors encountered. In addition the success/failure criteria performed well, correctly evaluating 95% of the automated actions."

Show More