Anomalous behavior identification using statistical analysis of large scale user interaction data

Abstract

Abstract: The challenge of identifying usability problems in interactive applications has been dealt with by companies for decades, but the amount of issues found in production systems illustrates how far we are from a widely usable solution. The integration of statistical analysis of large scale user interaction data into a user centered design process, presented by the authors in an earlier work [1], can significantly improve the chance of identifying usability problems in certain classes of applications. In this article, an expansion of the approach is proposed, leveraging the concept of ‘task’ as defined in the ISO 9241-11 [2] to create the basis for the automatic identification of anomalous interaction behavior. Here, ‘anomalous’ is understood as any statistically significant deviation from the expected interaction behavior, as defined in the implemented information architecture and navigation flow, or from the most often observed interaction pattern. With that, we argue, a relevant new tool to support the process of usability evaluation is created, uncovering interaction patterns not easily identifiable by other means