HackerPapers

arXiv:2209.07663cs.IR

Monolith: Real Time Recommendation System With Collisionless Embedding Table

Zhuoran Liu, Leqi Zou, Xuan Zou, Caihua Wang, Biao Zhang, Da Tang, Bolin Zhu, Yijie Zhu, Peng Wu, Ke Wang, Youlong Cheng

Abstract

Building a scalable and real-time recommendation system is vital for many businesses driven by time-sensitive customer feedback, such as short-videos ranking or online ads. Despite the ubiquitous adoption of production-scale deep learning frameworks like TensorFlow or PyTorch, these general-purpose frameworks fall short of business demands in recommendation scenarios for various reasons: on one hand, tweaking systems based on static parameters and dense computations for recommendation with dynamic and sparse features is detrimental to model quality; on the other hand, such frameworks are designed with batch-training stage and serving stage completely separated, preventing the model from interacting with customer feedback in real-time. These issues led us to reexamine traditional approaches and explore radically different design choices. In this paper, we present Monolith, a system tailored for online training. Our design has been driven by observations of our application workloads and production environment that reflects a marked departure from other recommendations systems. Our contributions are manifold: first, we crafted a collisionless embedding table with optimizations such as expirable embeddings and frequency filtering to reduce its memory footprint; second, we provide an production-ready online training architecture with high fault-tolerance; finally, we proved that system reliability could be traded-off for real-time learning. Monolith has successfully landed in the BytePlus Recommend product.

arXiv comments
ORSUM@ACM RecSys 2022
Journal reference
DOI
Subjects
cs.IRInformation Retrieval

On Hacker News

Linked by 10 identifiable commenters in 10 comments across 10 threads, from Dec 2022 to Feb 2026. Found in 9 HN submissions from Nov 2022 to Feb 2026. Open submission discussions.

Comments linking this paper, per year.

Discussed in comments10 threads

  1. TikTok's 'addictive design' found to be illegal in Europe 1 comment

    • I gave a talk at PyData Berlin on how to build your own TikTok recommendation algorithm. The TikTok personalized recommendation engine is the world's most valuable AI. It's TikTok's differentiation. It updates recommendations within 1 second of you clicking - at human…

      jamesblondeView on HNParent context

  2. You Don't Need Re-Ranking: Understanding the Superlinked Vector Layer 1 comment

    • If this was a solved problem, every e-com website would already run some variant of https://arxiv.org/pdf/2209.07663 for all their shopping surfaces and some version of "deep research" on every search query. They don't, so the problem isn't solved. And a big part of the problem…

      supoView on HNParent context

  3. TikTok preparing for U.S. shut-off on Sunday 1 comment

    • the tiktok algo is genuinely impressive. What's cool is that the engineers published some works explaining how it functions. https://arxiv.org/abs/2209.07663 It's an interesting read if you're into recommender systems or AI in general. What amazes me is that despite this…

      GrimblewaldView on HNParent context

  4. TikTok divestment law upheld by federal appeals court 1 comment

    • At best this comment implies there's actual an actual smoking gun related to what functionally are nation-state capability and operations to influence populations external to their own, specifically related to TikTok/ByteDance. If you have an AP government level understanding of…

      wyldberryView on HNParent context

  5. X is changing its algorithm to highlight smaller accounts 1 comment

    • If you think about it, the secret sauce tends to be in the data they're able to capture and input into the recommendation algorithms rather than the recommendation algorithm itself. I mean, even Tiktok has papers detailing their recommendation algorithm [1]. So does Youtube and…

      jerojeroView on HNParent context

  6. Monolith: The Recommendation System Behind TikTok 1 comment
  7. The secret sauce of TikTok’s recommendations 1 comment

    • Relate HN comments: - https://news.ycombinator.com/item?id=33494796 Related paper: - https://arxiv.org/abs/2209.07663 Related GitHub: - https://github.com/bytedance/monolith

      O__________OView on HNParent context

  8. Source code for Monolith, TikTok's recommendation system 1 comment
  9. Ask HN: Techwise, how did TikTok get so good? 1 comment

    • This paper by various folks at Bytedance hints at their recommender architecture: https://arxiv.org/pdf/2209.07663.pdf

      sgwizdakView on HNParent context

  10. TikTok’s Secret Sauce 1 comment

    • Arvind's making the case that the algorithm isn't particularly special and it's all about the system design which hooks users. It's definitely true that a huge part of their user retention isn't some magical algorithm, but they have made some genuine algorithmic/training…

      CharlieSweetingView on HNParent context

Submitted to Hacker News

Connected through shared HN threads, not inferred similarity or a recommendation.