Data-driven modelling and probabilistic analysis of interactive software usage

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