Building information systems using collaborative-filtering recommendation techniques

IoT-technologies allow for the connection of miscellaneous devices, thereby creating a platform that sustains rich data sources. Given the circumstances, it is essential to have decent machinery in order to exploit the existing infrastructure and provide users with personalized services. Among others, recommender systems have been widely used to suggest users additional items that best match their needs and expectation. The use of recommender systems has gained considerable momentum in recent years. Nevertheless, the selection of a proper recommendation technique depends much on the input data as well as the domain of applications. In this work, we present an evaluation of two well-known collaborative-filtering (CF) techniques to build an information system for managing and recommending books in the IoT context. To validate the performance, we conduct a series of experiments on two considerably large datasets. The experimental results lead us to some interesting conclusions. In contrast to many existing studies which state that the item-based CF technique outperforms the user-based CF technique, we found out that there is no distinct winner between them. Furthermore, we confirm that the performance of a CF recommender system may be good with regards to some quality metrics, but not to some others.