Analysing user behaviour through dynamic population models.

We apply a statistical modelling-based approach to exploring, analysing and predicting behavioural patterns of users of mobile software. The technique employed represents the behaviour of each user through a weighted mixture over data-generating distributions. In the described pilot study, we show how we have modelled the behaviour of over a hundred users of an iOS game. We illustrate how this modelling approach can be used to determine user play strategies and learning rates and show how this affects the length of time users keep returning to play the game. We describe our ongoing work, including feeding results of the modelling into the design process.