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
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:160330-160351, 2026.

Abstract

Large Language Models (LLMs) represent a promising frontier for recommender systems, yet their development has been impeded by the absence of predictable scaling laws, which are crucial for guiding research and optimizing resource allocation. We hypothesize that this may be attributed to the inherent noise, bias, and incompleteness of raw user interaction data in prior continual pre-training (CPT) efforts. This paper introduces a novel, layered framework for generating high-quality synthetic data that circumvents such issues by creating a curated, pedagogical curriculum for the LLM. We provide powerful, direct evidence for the utility of our curriculum by showing that standard sequential models trained on our principled synthetic data significantly outperform ($+130%$ on recall@100 for SasRec) models trained on real data in downstream ranking tasks, demonstrating its superiority for learning generalizable user preference patterns. Building on this, we empirically demonstrate, for the first time, robust power-law scaling for an LLM that is continually pre-trained on our high-quality, recommendation-specific data. Our experiments reveal consistent and predictable perplexity reduction across multiple synthetic data modalities. These findings establish a foundational methodology for reliable scaling LLM capabilities in the recommendation domain, thereby shifting the research focus from mitigating data deficiencies to leveraging high-quality, structured information.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-zhang26ih, title = {Principled Synthetic Data Enables the First Scaling Laws for {LLM}s in Recommendation}, author = {Zhang, Benyu and Zhang, Qiang and Cheng, Jianpeng and Chen, Hong-You and Wang, Qifei and Sun, Wei and Li, Shen and Li, Jia and Wu, Jiahao and Fan, Xiangjun and Yan, Hong}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {160330--160351}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/zhang26ih/zhang26ih.pdf}, url = {https://proceedings.mlr.press/v306/zhang26ih.html}, abstract = {Large Language Models (LLMs) represent a promising frontier for recommender systems, yet their development has been impeded by the absence of predictable scaling laws, which are crucial for guiding research and optimizing resource allocation. We hypothesize that this may be attributed to the inherent noise, bias, and incompleteness of raw user interaction data in prior continual pre-training (CPT) efforts. This paper introduces a novel, layered framework for generating high-quality synthetic data that circumvents such issues by creating a curated, pedagogical curriculum for the LLM. We provide powerful, direct evidence for the utility of our curriculum by showing that standard sequential models trained on our principled synthetic data significantly outperform ($+130%$ on recall@100 for SasRec) models trained on real data in downstream ranking tasks, demonstrating its superiority for learning generalizable user preference patterns. Building on this, we empirically demonstrate, for the first time, robust power-law scaling for an LLM that is continually pre-trained on our high-quality, recommendation-specific data. Our experiments reveal consistent and predictable perplexity reduction across multiple synthetic data modalities. These findings establish a foundational methodology for reliable scaling LLM capabilities in the recommendation domain, thereby shifting the research focus from mitigating data deficiencies to leveraging high-quality, structured information.} }
Endnote
%0 Conference Paper %T Principled Synthetic Data Enables the First Scaling Laws for LLMs in Recommendation %A Benyu Zhang %A Qiang Zhang %A Jianpeng Cheng %A Hong-You Chen %A Qifei Wang %A Wei Sun %A Shen Li %A Jia Li %A Jiahao Wu %A Xiangjun Fan %A Hong Yan %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-zhang26ih %I PMLR %P 160330--160351 %U https://proceedings.mlr.press/v306/zhang26ih.html %V 306 %X Large Language Models (LLMs) represent a promising frontier for recommender systems, yet their development has been impeded by the absence of predictable scaling laws, which are crucial for guiding research and optimizing resource allocation. We hypothesize that this may be attributed to the inherent noise, bias, and incompleteness of raw user interaction data in prior continual pre-training (CPT) efforts. This paper introduces a novel, layered framework for generating high-quality synthetic data that circumvents such issues by creating a curated, pedagogical curriculum for the LLM. We provide powerful, direct evidence for the utility of our curriculum by showing that standard sequential models trained on our principled synthetic data significantly outperform ($+130%$ on recall@100 for SasRec) models trained on real data in downstream ranking tasks, demonstrating its superiority for learning generalizable user preference patterns. Building on this, we empirically demonstrate, for the first time, robust power-law scaling for an LLM that is continually pre-trained on our high-quality, recommendation-specific data. Our experiments reveal consistent and predictable perplexity reduction across multiple synthetic data modalities. These findings establish a foundational methodology for reliable scaling LLM capabilities in the recommendation domain, thereby shifting the research focus from mitigating data deficiencies to leveraging high-quality, structured information.
APA
Zhang, B., Zhang, Q., Cheng, J., Chen, H., Wang, Q., Sun, W., Li, S., Li, J., Wu, J., Fan, X. & Yan, H.. (2026). Principled Synthetic Data Enables the First Scaling Laws for LLMs in Recommendation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:160330-160351 Available from https://proceedings.mlr.press/v306/zhang26ih.html.

Related Material