On component interactions in two-stage recommender systems
arXiv:2106.14979
Abstract
Thanks to their scalability, two-stage recommenders are used by many of today's largest online platforms, including YouTube, LinkedIn, and Pinterest. These systems produce recommendations in two steps: (i) multiple nominators, tuned for low prediction latency, preselect a small subset of candidates from the whole item pool; (ii) a slower but more accurate ranker further narrows down the nominated items, and serves to the user. Despite their popularity, the literature on two-stage recommenders is relatively scarce, and the algorithms are often treated as mere sums of their parts. Such treatment presupposes that the two-stage performance is explained by the behavior of the individual components in isolation. This is not the case: using synthetic and real-world data, we demonstrate that interactions between the ranker and the nominators substantially affect the overall performance. Motivated by these findings, we derive a generalization lower bound which shows that independent nominator training can lead to performance on par with uniformly random recommendations. We find that careful design of item pools, each assigned to a different nominator, alleviates these issues. As manual search for a good pool allocation is difficult, we propose to learn one instead using a Mixture-of-Experts based approach. This significantly improves both precision and recall at K.
arXiv commentAppears in the proceedings of the NeurIPS 2021 conference
References in corpus (22)
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- A Contextual-Bandit Approach to Personalized News Article Recommendation
- Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems
- Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity
- Hybrid Recommender Systems: A Systematic Literature Review
- Learning Tree-based Deep Model for Recommender Systems
- Doubly Robust Policy Evaluation and Learning
- Taming the Monster: A Fast and Simple Algorithm for Contextual Bandits
- Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
- Recommendations as Treatments: Debiasing Learning and Evaluation
- Efficient Optimal Learning for Contextual Bandits
- Learning from Logged Implicit Exploration Data
- On the Difficulty of Evaluating Baselines: A Study on Recommender Systems
- A Contextual Bandit Bake-off
- From Optimizing Engagement to Measuring Value
- Practical Contextual Bandits with Regression Oracles
- Recommendations and User Agency: The Reachability of Collaboratively-Filtered Information
- Adapting to Misspecification in Contextual Bandits
- Do Offline Metrics Predict Online Performance in Recommender Systems?
- Adaptive Trade-Offs in Off-Policy Learning
- Exploration in two-stage recommender systems
- Tractable contextual bandits beyond realizability