Look what you've done! Task recognition based on PC activities

days many people do their work on PCs. Due to interruptions and task switches it is quite common that at the end of a working day these people experience a feeling of having lost track of what they have been doing during the day. In this thesis a tool was developed which automatically recognizes the tasks a user is performing and presents these in an overview. It was investigated which task labels humans intuitively use and in how far it is possible to recognize these tasks on the basis of low level computer activity, like used applications and clicking and typing behavior. Results show that after only a few hours of training a reasonable classification accuracy can be reached. Comparison of several classification approaches reveals that there is not one classifier that suits all users best in terms of classification performance and learning Individual differences, due to mix of performed tasks and the individual work style, indicate that the tool should be personalized.