Automating Management of Resources on Desktop Machines

Two challenges that affect retrieval tasks in a typical computer desktop user environment are the speed and accuracy of accessing information such as documents and media files. Generating a common model to fit patterns of use with a reasonable level of prediction is challenging. This is as the individual usage patterns on the desktop is assumed to vary largely depending on the context of task. In this paper, we propose a lightweight learning framework, Usage Provenance and Prediction (UPP) model, which traces the user's mouse input and predicts the associated application with the media content. We disseminate the architecture of our tracing framework that is embedded on modern operating systems and show that this model adapts and improves data seeking experience at a user level. Our UPP framework achieves 96.70% accuracy with average application speed improvements of up to 32.92% over standard application launches. The results are encouraging and supports the potential of automating resources through the application of tracing usage patterns.