Recommendations beyond the ratings matrix

Abstract

Recommender systems have become indispensable for several Web sites, such as Amazon, Netflix and Google News, helping users navigate through the abundance of available choices. Although the field has advanced impressively in the last years with respect to models, usage of heterogeneous information, such as ratings and text reviews, and recommendations for modern applications beyond purchases, almost all of the approaches rely on the data that exist within the recommender and on user explicit input. In a rapidly connected world, though, information is not isolated and does not necessarily lie in the database of a single recommender. Rather, Web offers tremendous amount of information on almost everything, from items to users and their tendency to certain items, but also information on general trends and demographics. We envision an out-of-the-box recommender system that exploits the existing information in a recommender, namely, items, users and ratings, but also explores new sources of information out of the database, like user online traces and online discussions about data items, and exploits them for better and innovative recommendations. We discuss the challenges that such an out-of-the-box approach effects and how it reshapes the field of recommenders.