IoT-CR: A Novel IoT-Based Approach to Address Data Sparsity in Context-Aware Recommendation Systems

Abstract

In recent years, recommendation systems (RS) have become an essential part of modern online services, helping users discover new products, content, and experiences that match their preferences and needs. In particular, context-aware recommendation systems (CARS) have received considerable attention because of their capacity to use contextual information to deliver relevant and personalized recommendations, compared to traditional recommendation systems that only use user preferences and item attributes to make recommendations. However, the performance and effectiveness of CARS are challenged by the rise of data sparsity, a common issue in many recommender systems. It occurs when there is an insufficient amount of user-item interactions. This study explores using varied IoT contextual data to address this issue. We present and assess an IoT-assisted Contextual Recommendation (IoT-CR) system, which is an end-to-end deep learning framework architecture that aims to incorporate rich contexts from IoT sensors seamlessly into the recommendation process. To prove this concept, we perform an extensive comparative study against a set of baseline models on four different public, context-rich datasets. We find a mixed bag of results, indicating that the performance of models depends very much on the characteristics of the datasets such as their size and sparsity. In particular, the IoT-CR framework achieves the best results on the largest dataset where there is enough data for it to learn complex interactions. On the other hand, in smaller or more sparse situations, classical collaborative filtering or tree-based models perform better. This research offers an important benchmark, stating that although supplementing data with IoT signals is a very good way forward, the effectiveness of complex models is not a general case and depends critically on the data landscape.