The recommender problem revisited

In 2006, Netflix announced a $1M prize competition to advance recommendation algorithms. The recommendation problem was simplified as the accuracy in predicting a user rating measured by the Root Mean Squared Error (RMSE). While that formulation helped get the attention of the research community in the area, it may have put an excessive focus on what is simply one of the many possible approaches to recommendations. In this tutorial we will describe different components of modern recommender systems such as: personalized ranking, similarity, explanations, context-awareness, or search as recommendation. We will use the Netflix use case as a driving example of a prototypical industrial-scale recommender system that has evolved from focusing on rating prediction to full page optimization. We will also review the usage of modern algorithmic approaches that include algorithms such as Factorization Machines [9], Restricted Boltzmann Machines [10], SimRank [7], Deep Neural Networks, or Listwise Learning-to-rank [6, 12, 11]. The original recommendation problem was formulated around the existence of explicit user ratings. However, recommender systems can be built using different kinds of data including implicit behavioral data, social connections, or demographics. In this tutorial we will review the usage of different data types and discuss what the availability of Big Data has brought into the research area. Finally, and also related to the availability of large quantities of data, we will talk about how system and architectural decisions play a role in understanding the recommender problem. This tutorial is in part based on recent publications by the author [5, 1, 8, 2, 4, 3].