Modeling task deviations as eccentricity distribution peaks
Detailed usage data is becoming available through different devices (e.g., personal computer, cell phones, tablets, watches, glasses, wrist bands), in huge volumes, and in a speed that requires new models and visualizations to support the understanding of detailed user actions at scale. Without appropriate methods that summarize or provide means of analyzing large usage data sets, a semantic gap between the event-by-event data and the tasks profile remains. In this context, this work proposes a technique to support the analysis of task deviation from the examination of detailed user interface events streams. From the analysis of 427 event-by-event logged sessions (captured under user consent) of a technical reference website, this work presents how to identify task deviations by using eccentricity distribution. The proposed technique is a promising way of identifying task deviations in large log data sets containing information about how users performed real tasks.