Improving the Cold-Start Problem in User Task Automation by Using Models at Runtime

Abstract

Automating user behaviour is one of the most important challenges in Ambient Intelligence. Most of the proposed approaches to confront this are based on machine-learning algorithms. Using them, user routine tasks are inferred from user past actions and then automated when needed. However, these approaches present the cold-start problem, i.e. they cannot infer routine tasks until they gather sufficient user actions. We improve this problem by using a modelling approach. We propose two models to specify the routine tasks known at design time and a software infrastructure that automates them when needed by interpreting these models at runtime. Thus, we achieve to automate user routine tasks from the very beginning. Furthermore, we complement this infrastructure with a model-based API that allows these models to be modified at runtime. Thus, we provide a high-level repository for user behaviour information and an initial background to improve behaviour predictions.