Latent workspaces improve long-sequence recommendations with efficient reasoning
Read More, Reason Deeper: Latent Workspaces for Long-Sequence Recommendation
Information Retrieval
Summary
Recommending items based on long histories is tricky because systems mix how much data they look at with how deeply they analyze it. The authors introduce a new approach called Latent Workspace Ranker that keeps evidence separate and lets the system look at more data and analyze it more deeply without slowing down too much. This separation allows recommendations to improve by using longer history and more thinking steps independently. Their method shows better results and efficiency on very large datasets compared to existing techniques.
What this means in practice
- •For e-commerce teams: Improve product recommendations by efficiently using very long user behavior histories and deeper reasoning steps to boost ranking accuracy while reducing compute cost.
- •For video streaming platforms: Enhance content ranking quality on large catalogs by independently scaling history length and reasoning depth without excessive computational resources.
Authors
Renzhi Wu, Li Sheng, Zikun Cui, Junjie Yang, Tai Guo, Keke Zhai, Yan Li, Jerry Fu, Nipun Mathur, Sri Reddy, Hong Li
Abstract
Long-sequence recommendation can improve along two axes: \emph{evidence scaling} exposes longer histories, while \emph{reasoning scaling} adds layers that interpret that evidence for a candidate. Existing models couple them: sequence encoders propagate every added event through depth, whereas one-shot selection prevents later layers from revisiting omitted evidence. We introduce the Latent Workspace Ranker (LWR), which keeps typed evidence in persistent banks and propagates a fixed-size candidate-local workspace. Each layer rereads the banks from its evolving workspace and archives compact post-interaction state, allowing history length and reasoning depth to grow independently. We compare four architectures across various history--depth configurations using evidence scaling at fixed depth, reasoning scaling at fixed 10K history, and their joint compute-efficient frontier. LWR leads all families at long histories, continues improving with reasoning compute, and converts additional training compute into quality gains at the fastest fitted rate on the joint frontier. At 10K history, it improves NE by 0.19% over HSTU while using 69.6% fewer forward GFLOP/request. It also outperforms existing baselines on two billion-scale ranking surfaces. These results show that separating addressable evidence from propagated state makes reading more and reasoning deeper independently controllable and jointly useful.