Copertina

OneRec-V2 Technical Report

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Zhou, Guorui, Hu, Hengrui, Cheng, Hongtao, Wang, Huanjie, Deng, Jiaxin, Zhang, Jinghao, Cai, Kuo, Ren, Lejian, Ren, Lu, Yu, Liao, Zheng, Pengfei, Luo, Qiang, Wang, Qianqian, Hu, Qigen, Huang, Rui, Tang, Ruiming, Wang, Shiyao, Yang, Shujie, Wu, Tao, Li, Wuchao, Luo, Xinchen, Wang, Xingmei, Su, Yi, Wu, Yunfan, Cheng, Zexuan, Liu, Zhanyu, Zhang, Zixing, Zhang, Bin, Wang, Boxuan, Ma, Chaoyi, Song, Chengru, Wang, Chenhui, Chu, Chenglong, Wang, Di, Meng, Dongxue, Zang, Dunju, Yang, Fan, Zhang, Fangyu, Jiang, Feng, Zhang, Fuxing, Wang, Gang, Zhang, Guowang, Li, Han, Bao, Honghui, Cao, Hongyang, Huang, Jiaming, Chen, Jiapeng, Liu, Jiaqiang, Jia, Jinghui, Gai, Kun, Hu, Lantao, Zeng, Liang, Wang, Qiang, Zhou, Qidong, Zhang, Rongzhou, Wang, Shengzhe, He, Shihui, Yang, Shuang, Mao, Siyang, Huang, Sui, He, Tiantian, Gao, Tingting, Yuan, Wei, Liang, Xiao, Xu, Xiaoxiao, Liu, Xugang, Wang, Yan, Zhou, Yang, Wang, Yi, Liu, Yiwu, Song, Yue, Zhang, Yufei, Zhao, Yunfeng, Ling, Zhixin, Li, Ziming
Natura:	Preprint
Pubblicazione:	2025
Soggetti:	Information Retrieval
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

_version_	1866912673343471616
author	Zhou, Guorui Hu, Hengrui Cheng, Hongtao Wang, Huanjie Deng, Jiaxin Zhang, Jinghao Cai, Kuo Ren, Lejian Ren, Lu Yu, Liao Zheng, Pengfei Luo, Qiang Wang, Qianqian Hu, Qigen Huang, Rui Tang, Ruiming Wang, Shiyao Yang, Shujie Wu, Tao Li, Wuchao Luo, Xinchen Wang, Xingmei Su, Yi Wu, Yunfan Cheng, Zexuan Liu, Zhanyu Zhang, Zixing Zhang, Bin Wang, Boxuan Ma, Chaoyi Song, Chengru Wang, Chenhui Chu, Chenglong Wang, Di Meng, Dongxue Zang, Dunju Yang, Fan Zhang, Fangyu Jiang, Feng Zhang, Fuxing Wang, Gang Zhang, Guowang Li, Han Bao, Honghui Cao, Hongyang Huang, Jiaming Chen, Jiapeng Liu, Jiaqiang Jia, Jinghui Gai, Kun Hu, Lantao Zeng, Liang Wang, Qiang Zhou, Qidong Zhang, Rongzhou Wang, Shengzhe He, Shihui Yang, Shuang Mao, Siyang Huang, Sui He, Tiantian Gao, Tingting Yuan, Wei Liang, Xiao Xu, Xiaoxiao Liu, Xugang Wang, Yan Zhou, Yang Wang, Yi Liu, Yiwu Song, Yue Zhang, Yufei Zhao, Yunfeng Ling, Zhixin Li, Ziming
author_facet	Zhou, Guorui Hu, Hengrui Cheng, Hongtao Wang, Huanjie Deng, Jiaxin Zhang, Jinghao Cai, Kuo Ren, Lejian Ren, Lu Yu, Liao Zheng, Pengfei Luo, Qiang Wang, Qianqian Hu, Qigen Huang, Rui Tang, Ruiming Wang, Shiyao Yang, Shujie Wu, Tao Li, Wuchao Luo, Xinchen Wang, Xingmei Su, Yi Wu, Yunfan Cheng, Zexuan Liu, Zhanyu Zhang, Zixing Zhang, Bin Wang, Boxuan Ma, Chaoyi Song, Chengru Wang, Chenhui Chu, Chenglong Wang, Di Meng, Dongxue Zang, Dunju Yang, Fan Zhang, Fangyu Jiang, Feng Zhang, Fuxing Wang, Gang Zhang, Guowang Li, Han Bao, Honghui Cao, Hongyang Huang, Jiaming Chen, Jiapeng Liu, Jiaqiang Jia, Jinghui Gai, Kun Hu, Lantao Zeng, Liang Wang, Qiang Zhou, Qidong Zhang, Rongzhou Wang, Shengzhe He, Shihui Yang, Shuang Mao, Siyang Huang, Sui He, Tiantian Gao, Tingting Yuan, Wei Liang, Xiao Xu, Xiaoxiao Liu, Xugang Wang, Yan Zhou, Yang Wang, Yi Liu, Yiwu Song, Yue Zhang, Yufei Zhao, Yunfeng Ling, Zhixin Li, Ziming
contents	Recent breakthroughs in generative AI have transformed recommender systems through end-to-end generation. OneRec reformulates recommendation as an autoregressive generation task, achieving high Model FLOPs Utilization. While OneRec-V1 has shown significant empirical success in real-world deployment, two critical challenges hinder its scalability and performance: (1) inefficient computational allocation where 97.66% of resources are consumed by sequence encoding rather than generation, and (2) limitations in reinforcement learning relying solely on reward models. To address these challenges, we propose OneRec-V2, featuring: (1) Lazy Decoder-Only Architecture: Eliminates encoder bottlenecks, reducing total computation by 94% and training resources by 90%, enabling successful scaling to 8B parameters. (2) Preference Alignment with Real-World User Interactions: Incorporates Duration-Aware Reward Shaping and Adaptive Ratio Clipping to better align with user preferences using real-world feedback. Extensive A/B tests on Kuaishou demonstrate OneRec-V2's effectiveness, improving App Stay Time by 0.467%/0.741% while balancing multi-objective recommendations. This work advances generative recommendation scalability and alignment with real-world feedback, representing a step forward in the development of end-to-end recommender systems.
format	Preprint
id	arxiv_https___arxiv_org_abs_2508_20900
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	OneRec-V2 Technical Report Zhou, Guorui Hu, Hengrui Cheng, Hongtao Wang, Huanjie Deng, Jiaxin Zhang, Jinghao Cai, Kuo Ren, Lejian Ren, Lu Yu, Liao Zheng, Pengfei Luo, Qiang Wang, Qianqian Hu, Qigen Huang, Rui Tang, Ruiming Wang, Shiyao Yang, Shujie Wu, Tao Li, Wuchao Luo, Xinchen Wang, Xingmei Su, Yi Wu, Yunfan Cheng, Zexuan Liu, Zhanyu Zhang, Zixing Zhang, Bin Wang, Boxuan Ma, Chaoyi Song, Chengru Wang, Chenhui Chu, Chenglong Wang, Di Meng, Dongxue Zang, Dunju Yang, Fan Zhang, Fangyu Jiang, Feng Zhang, Fuxing Wang, Gang Zhang, Guowang Li, Han Bao, Honghui Cao, Hongyang Huang, Jiaming Chen, Jiapeng Liu, Jiaqiang Jia, Jinghui Gai, Kun Hu, Lantao Zeng, Liang Wang, Qiang Zhou, Qidong Zhang, Rongzhou Wang, Shengzhe He, Shihui Yang, Shuang Mao, Siyang Huang, Sui He, Tiantian Gao, Tingting Yuan, Wei Liang, Xiao Xu, Xiaoxiao Liu, Xugang Wang, Yan Zhou, Yang Wang, Yi Liu, Yiwu Song, Yue Zhang, Yufei Zhao, Yunfeng Ling, Zhixin Li, Ziming Information Retrieval Recent breakthroughs in generative AI have transformed recommender systems through end-to-end generation. OneRec reformulates recommendation as an autoregressive generation task, achieving high Model FLOPs Utilization. While OneRec-V1 has shown significant empirical success in real-world deployment, two critical challenges hinder its scalability and performance: (1) inefficient computational allocation where 97.66% of resources are consumed by sequence encoding rather than generation, and (2) limitations in reinforcement learning relying solely on reward models. To address these challenges, we propose OneRec-V2, featuring: (1) Lazy Decoder-Only Architecture: Eliminates encoder bottlenecks, reducing total computation by 94% and training resources by 90%, enabling successful scaling to 8B parameters. (2) Preference Alignment with Real-World User Interactions: Incorporates Duration-Aware Reward Shaping and Adaptive Ratio Clipping to better align with user preferences using real-world feedback. Extensive A/B tests on Kuaishou demonstrate OneRec-V2's effectiveness, improving App Stay Time by 0.467%/0.741% while balancing multi-objective recommendations. This work advances generative recommendation scalability and alignment with real-world feedback, representing a step forward in the development of end-to-end recommender systems.
title	OneRec-V2 Technical Report
topic	Information Retrieval
url	https://arxiv.org/abs/2508.20900