Deep Neural Networks for YouTube Recommendations
By Paul Covington, Jay K. Adams, Emre Sargin. 2016
3,933 citations recorded by the source corpus.
YouTube represents one of the largest scale and most sophisticated industrial recommendation systems in existence. In this paper, we describe the system at a high level and focus on the dramatic performance improvements brought by deep learning. The paper is split according to the classic two-stage information retrieval dichotomy: first, we detail a deep candidate generation model and then describe a separate deep ranking model. We also provide practical lessons and insights derived from designing, iterating and maintaining a massive recommendation system with enormous user-facing impact.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Information Systems, Large Language Models, Recommender Systems
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Sergey Levine
- Yoshua Bengio
- Percy Liang
- Mohit Bansal
- Quoc V. Le
- Jianfeng Gao
- Xiangnan He
- Kyunghyun Cho
In Computer Vision and Pattern Recognition
- Dacheng Tao
- Luc Van Gool
- Trevor Darrell
- Yi Yang
- Xiaogang Wang
- Ming-Hsuan Yang
- Jitendra Malik
- Andrew Zisserman
In Information Systems
In Large Language Models
In Recommender Systems
Citations in the corpus
- Learning to summarize from human feedback (2020, background): “This can lead to problems, such as Youtube recommendations promoting click-bait [11].”
- Degenerate Feedback Loops in Recommender Systems (2019, background): “Recommender systems are increasingly used to provide users with personalized product and information offerings (Ben Schafer, Konstan, and Riedl, 2001; Lu et al., 2015; Covington, Adams, and Sargin, 2016).”
- Variational Autoencoders for Collaborative Filtering (2018, methodology): “For example, it has been used in recurrent neural networks for session-based sequential recommendation [8, 15, 16, 42, 45] and in feedward neural networks applied to Youtube recommendation [9].”
- Graph Convolutional Neural Networks for Web-Scale Recommender Systems (2018, background): “We also do not consider non-deep learning approaches for generating item/content embeddings, since other works have already proven state-of-the-art performance of deep learning approaches for generating such embeddings [9, 12, 24].”
- DKN: Deep Knowledge-Aware Network for News Recommendation (2018, methodology): “In addition to the aforementioned DSSM [16], DeepWide [6], DeepFM [13], YouTubeNet [8] and DMF [50], other popular deep-learning-based recommender systems include Collaborative Deep Learning [44], SHINE [45], Multi-view Deep Learning [11], and Neural Collaborative Filtering [14].”
- Deep Interest Network for Click-Through Rate Prediction (2017, methodology): “Most of the popular model structures [3, 4, 21] share a similar Embedding&MLP paradigm, which we refer to as base model, as shown in the left of Fig.”
- Explainable Recommendation: A Survey and New Perspectives (2018, methodology): “is also applied in real-world commercial systems such as the YouTube recommender system (Covington et al., 2016).”
- Differentiable Neural Input Search for Recommender Systems (2020, background): “Another notable fact is that embedding layers in industrial web-scale recommender systems [10, 31] account for the majority of model parameters and can consume hundreds of gigabytes of memory space.”
- Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift (2018, background): “Software systems employing deep neural networks are now applied widely in industry, powering the vision systems in social networks [47] and self-driving cars [5], providing assistance to radiologists [24], underpinning recommendation engines used by online platforms [9, 12], enabling the bestperform”
- MOReL : Model-Based Offline Reinforcement Learning (2020, background): “[12], and has applications in healthcare [33, 34, 35], recommendation systems [36, 37, 38, 39], dialogue systems [40, 19, 41], and autonomous driving [42].”
- Behavior Regularized Offline Reinforcement Learning (2019, background): “For example, for recommendation systems (Li et al., 2011; Covington et al., 2016) or health applications (Murphy et al., 2001), deploying a new policy may only be done at a low frequency after extensive testing and evaluation.”
- Challenges of Real-World Reinforcement Learning (2019, background): “This may be on the order of milliseconds for a recommender system (Covington et al., 2016) responding to a user request or the control of a physical robot, and up to the order of minutes for building control systems (Evans & Gao).”
- Are Labels Required for Improving Adversarial Robustness? (2019, methodology): “System based on these techniques are now being developed and deployed for a wide variety of applications, from recommending/ranking content on the web [13, 22] to autonomous driving [7] and even in medical diagnostics [14].”
- Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts (2018, methodology): “In [11], a shared-bottom model is used to learn a ranking algorithm for video recommendation.”
- DAPPLE: a pipelined data parallel approach for training large models (2020, background): “, NLP[39], search/recommendation systems[13, 45]) have billions of parameters, demanding tens to hundreds of GBs of device memory for training.”
- Neural Input Search for Large Scale Recommendation Models (2019, methodology): “However, the input component, although contains a great portion of model parameters due to large-sized embeddings, has been frequently designed heuristically across industry, such as YouTube ([7]), Google Play ([6]), Netflix ([9]), etc.”
- Information Leakage in Embedding Models (2020, background): “Embeddings have also been successfully applied to other data domains including social networks [21], source code [2], YouTube watches [11], movie feedback [25], locations [12], etc.”