Datasets for Context-Aware Recommender Systems: Current Context and Possible Directions

Recommender systems alleviate the overload experienced by users when they have to choose among a large set of options. They are interesting both from a commercial point of view, to propose items to potential customers who may find them relevant, as well as from the point of view of end users, who will be more satisfied if they can find what they need in a short time and without having to navigate and process a large amount of data. Popular recommenders are those used by companies such as Amazon, Netflix or Pandora, to cite a few examples. It has been stated that context data could be exploited in order to provide more relevant recommendations to mobile users. Based on this idea, context-aware recommender systems have attracted significant research attention in the last years. However, in this position paper, we argue that the number of available datasets with context data is scarce and that even those datasets that incorporate context data are usually too sparse. Moreover, existing datasets focus on specific use cases that may not correspond to the one we need to consider for evaluation. Therefore, we analyze possible alternative and future directions that could be followed to mitigate this data availability problem.