Behaviour mining for automatic task-keeping and visualisations for task-refinding

Abstract

When people perform some tasks on their desktop, they tend to spend a considerable amount of time looking back, establishing past references and remembering. A task contains an evolving collection of documents to which the user is referring and that are relevant to the task. In the case of Web documents people tend to rely on their organisational skills and use a variety of “keeping” and “re-finding” tools which traditionally include bookmarks, history and search, amongst others. This paper reports the results of surveys conducted to investigate the way that people keep and refind Web based tasks. As tasks can be interrupted or con- ducted over different sessions a user needs to keep, re-find, and resume them. A prototype called PiMxT is also presented, which uses an incremental density-based clustering approach to automatically generate and keep task-clusters by considering the user’s window and tab switching, and re-visitation behaviour. The algorithm attempts to identify those documents that pertain to the same task-cluster, and also when a switch between two documents is effectively a task-switch. PiMxT allows for searching and filtering across the task-clusters as well as task-resumption through different visualisations. This tool was evaluated through a usability study and a questionnaire, and it was found that the time taken by the 15 participants to re-find and resume tasks was on average 44% less in 77% of the cases. Furthermore, although the participants expected PiMxT to be more complete and accurate they all recognised its usefulness.