Learning and Predicting User Behavior for Particular Resource Use

To successfully interact with users in providing useful information, intelligent user interfaces need a mechanism for recognizing, characterizing, and predicting user actions. In particular, it is our interest to develop the mechanism for recognizing and predicting simple user intentions, i.e., an activity involves in using particular resources. Much work to date in adaptive user interfaces has resulted in ad-hoc approaches such as simply capturing user preferences at a shallow level ignoring the more difficult problem of capturing the user intention. We frame the modeling task of user interface systems in terms of learning user patterns of using particular resources by understanding temporal information of activity, user intentions, and abstraction of user behavior. Our approach learns the individual user models through time-series action analysis and abstraction. After capturing the dynamics of user behavior into regularities of user behavior(patterns), probabilistic user models are constructed to facilitate the predictions of resource usage with a sequence of currently observed actions in the Unix domain.