Discovering User Behavior Patterns in Personalized Interface Agents

Abstract

In human-computer interaction, user interface events can be recorded and organized into sequences of episodes. By computing their implication networks, episode frequencies, and some heuristic measures of interestingness, we can readily derive some application-specific episode association rules. In order to demonstrate the proposal method, we have developed a personalized interface agent that can take into consideration interface events in analyzing user goals. It can then delegate on behalf of the user to interact with the software based on the recognized plans. In order to adapt to different users’ needs, the agent can personalize its assistance by learning user profiles. Currently, we have used the Microsoft Word as a test case. By detecting and analyzing the patterns of user behavior in using Word, the agent can automatically assist the users in certain Word tasks. The pattern association can be achieved at several levels, i.e., text-level (phrase association), paragraphlevel (formatting association), and document-level (style and source association).