Data-driven modelling and probabilistic analysis of interactive software usage

This paper answers the research question: how can we model and understand the ways in which users actually interact with software, given that usage styles vary from user to user, and even from use to use for an individual user. Our first contribution is to introduce two new probabilistic, admixture models, inferred from sets of logged user traces, which include observed and latent states. The models encapsulate the temporal and stochastic aspects of usage, the heterogeneous and dynamic nature of users, and the temporal aspects of the time interval over which the data was collected (e.g. one day, one month, etc.). A key concept is activity patterns, which encapsulate common observed temporal behaviours shared across a set of logged user traces. Each activity pattern is a discrete-time Markov chain in which observed variables label the states; latent states specify the activity patterns. The second contribution is how we use parametrised, probabilistic, temporal logic properties to reason about hypothesised behaviours within an activity pattern, and between activity patterns. Different combinations of inferred model and hypothesised property afford a rich set of techniques for understanding software usage. The third contribution is a demonstration of the models and temporal logic properties by application to user traces from a software application that has been used by tens of thousands of users worldwide