Exploring Contextual Paradigms in Context-Aware Recommendations

Traditional recommendation systems utilise past users' preferences to predict unknown ratings and recommend unseen items. However, as the number of choices from content providers increases, additional information, such as context, has to be included in the recommendation process to improve users' satisfaction. Context-aware recommendation systems exploit the users' contextual information (e.g., location, mood, company, etc.) using three main paradigms: contextual pre-filtering, contextual post-filtering, and contextual modelling. In this work, we explore these three ways of incorporating context in the recommendation pipeline, and compare them on context-aware datasets with different characteristics. The experimental evaluation showed that contextual pre-filtering and contextual modelling yield similar performance, while the post-filtering approach achieved poorer accuracy, emphasising the importance of context in producing good recommendations.