Designing context models for CARS incorporating partially observable context

Abstract

Context modelling and context reasoning are widely used topics in Context-Aware Recommender Systems research. Based on our research, the development of context models in context-aware recommender systems is problematic in that many domain specific and application specific context models are developed with limited or no reuse and sharing capabilities. Furthermore, context-aware recommender systems that follow the representational view of context, design and model the fully observable context that is known at recommendation time but do not consider partially observable context with unknown values at recommendation time, that can nevertheless enhance the recommendation outcome. In this paper we deal with the above two issues by proposing a CARS design system that enables developers: (i) to easily and effectively design context models by defining, sharing and reusing context parameters and (ii) to utilize partially observable context at recommendation time by using an interactional approach that incorporates user feedback by applying a utility based algorithm on context models. © Springer International Publishing Switzerland 2015.