TreasureX Researchcs.LGMutable Sketches: Real-Time Recommendations Without Retraining
Hector J. Garcia, Nick Clayton · July 16, 2026
A new paper from researchers proposes "mutable sketches" to solve the embedding staleness problem in two-stage recommendation systems. Traditionally, when a user rates a new item, their embedding (a numerical representation of preferences) stays frozen until the next retrain cycle, causing outdated recommendations. The mutable sketch uses a KP-tree (a sparse segment tree with sum aggregation) to store user preferences, fits a low-rank projection once, and recomputes embeddings on-the-fly as new ratings arrive. The authors prove that each new observation tightens the prediction error envelope, a guarantee missing in methods like FunkSVD and eALS. On the KuaiRec dataset, the mutable sketch achieves 0.810 RMSE using only 1.8% of the data, compared to ALS's 0.822 RMSE using 100% of the data, and is 8x faster per batch. A new user receives personalized recommendations in under 1 millisecond after their first rating, with no model retraining. The KP-tree's norm-proportional sampling improves item coverage by 40-130% on sparse data, while uniform sampling suffices for dense matrices. This approach reduces computational costs and enables real-time personalization, making it attractive for e-commerce and content platforms.
Read more: https://arxiv.org/abs/2607.15242
Educational content only, generated with AI assistance from public research abstracts. Not investment advice. Thank you to arXiv for use of its open access interoperability.