Content area
Full text
Abstract
Deep learning has profoundly impacted many areas of machine learning. However, it took a while for its impact to be felt in the field of recommender systems. In this article, we outline some of the challenges encountered and lessons learned in using deep learning for recommender systems at Netflix. We first provide an overview of the various recommendation tasks on the Netflix service. We found that different model architectures excel at different tasks. Even though many deep-learning models can be understood as extensions of existing (simple) recommendation algorithms, we initially did not observe significant improvements in performance over well-tuned non-deep-learning approaches. Only when we added numerous features of heterogeneous types to the input data, deep-learning models did start to shine in our setting. We also observed that deep-learning methods can exacerbate the problem of offline-online metric (mis-)alignment. After addressing these challenges, deep learning has ultimately resulted in large improvements to our recommendations as measured by both offline and online metrics. On the practical side, integrating deep-learning toolboxes in our system has made it faster and easier to implement and experiment with both deeplearning and non-deep-learning approaches for various recommendation tasks. We conclude this article by summarizing our take-aways that may generalize to other applications beyond Netflix.
INTRODUCTION
In the early 2010s, deep learning was taking off in the machine-learning community fueled by impressive results on a variety of tasks in different domains including computer vision, speech recognition, and natural language processing (NLP). At that time there was a stir in the air within the recommender-systems research community: Will the wave of deep learning also wash over recommenders to deliver tremendous improvements? As with many others, we at Netflix were intrigued by this question and the potential of deep learning to improve our recommendations. While the answer is now quite clear that deep learning is useful for recommender systems, the path to understand where deep learning is beneficial over existing recommendation approaches was an arduous one. This is evidenced by how many years it took for such methods to get traction in the research community. But it was a rewarding path as evidenced by a subsequent bloom of work on the subject. Our own investigations into deep learning at Netflix took a similar path:...