AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Principled Synthetic Data Enables the First Scaling Laws for LLMs in Recommendation

Benyu Zhang, Qiang Zhang, Jianpeng Cheng, Hong-You Chen, Qifei wang, Wei Sun, Shen Li, Jia Li, Jiahao Wu, Xiangjun Fan, Hong Yan

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train sequential recommenders on synthetic UIH instead of real logs (TSTR vs TRTR)GRU4Rec, NARM, STAMP, SASRec · Merrec test, Recall@10/100/1000Recall@K: +130% on recall@100 for SasRecImproved
add collaborative filtering data alongside UIH data (ablation)Qwen3-4B CPT · UIH Full (avg)asymptotic loss L_inf: 0.95 -> 0.66Improved
train on UIH data alone and evaluate on CF (ablation)Qwen3-4B CPT · CF Both Unseenscaling exponent alpha: alpha ≈ 0No clear effect
include CF data in the mixture (ablation)Qwen3-4B CPT · CF Both Unseenasymptotic loss L_inf: 1.30 -> 0.35Improved
include item-text data in the mixture (ablation)Qwen3-4B CPT · Item-Textasymptotic loss L_inf: ≈3 -> ≈1.2Improved
include item-text data in the mixture (ablation)Qwen3-4B CPT · UIH Full (avg)asymptotic loss L_inf: 0.66 -> 0.76Hurt
raise UIH mixture ratio so reduced UIH data repeats about 16 times or more (ablation)Qwen3-4B CPT · held-out UIH perplexityperplexityHurt
scale model size from 0.6B to 8B with fixed 163B tokens (ablation)Qwen3 CPT · UIH OODperplexity: 2.01 -> 2.10Hurt
scale model size from 0.6B to 4B and beyond (ablation)Qwen3 CPT · Collaborative filteringperplexity: saturates at ~1.47 PPL at 4BNo clear effect
scale model size from 0.6B to 8B (ablation)Qwen3 CPT · General domainscaling exponent alpha: alpha 0.02-0.03No clear effect
scale model size from 0.6B to 8B (ablation)Qwen3 CPT · UIHscaling exponent alpha: 0.45 -> 0.59Improved
apply method to Gemma 3 family on proprietary social media datasetLayered synthetic CPT · Internal datasetscaling-law fit RMSE: RMSE < 0.035Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy