The challenge of preparational behaviours in preference learning for ubiquitous systems

Many ubiquitous computing systems employ intelligent components that learn how to adapt the user's environment on their behalf, by observing how the user has adapted such environments in the past. Such components employ monitoring and machine learning techniques to capture human behaviours and process them to extract adaptation rules (or user preferences). However, learning preferences from observations of behaviour introduces challenges that are not so compounded in other machine learning problem domains. One key issue is preparational behaviours (or pre-actions) which current preference learning solutions can struggle to handle. This paper uses pre-actions as an example discussion point and raises the question of whether preference learning solutions should take advantage of temporal data from real-world environments to improve performance. The key contribution of this paper is the introduction and analysis of a novel machine learning technique (the DIANNE) that utilises temporal data to handle user behaviour anomalies such as pre-actions.