UMass Machine Learning and Friends Lunch | Main / Task Tracer Toward An Activity-Oriented Desktop Interface

Knowledge workers are multi-taskers. Their work lives can be divided into multiple on-going projects or activities, and their time at the desktop interleaves work on these projects and activities. However, existing desktop user interfaces do not have any notion of coherent projects or activities. The TaskTracer system seeks to support these workers by organizing the files, folders, contact information, calendar appointments, and web sites (collectively known as "resources") according to the activities that they support. To use TaskTracer, the user defines a hierarchy of projects/activities and declares to TaskTracer what current task he/she is working on at each point in time. TaskTracer instruments Microsoft Windows and Office to gather data on the resources that are accessed by the user and associates them with the currently-declared task. It then provides project-related assistance through (a) the TaskExplorer (which makes it easy for the user to return to previously-accessed resources), (b) the FolderPredictor (which predicts the relevant folder for Open and SaveAs actions), and TaskNotes (which provides a task-related notebook). To reduce the need for the user to declare the current activity, we apply machine learning methods to predict the current activity of the user based on incoming email messages and desktop behavior.