APAO: Adaptive Prefix-Aware Optimization for Generative Recommendation
Yuanqing Yu, Yifan Wang, Weizhi Ma, Zhiqiang Guo, Min Zhang · 2026
Read APAO: Adaptive Prefix-Aware Optimization for Generative Recommendation by Yuanqing Yu, Yifan Wang, Weizhi Ma, Zhiqiang Guo, Min Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-02730.txt.
Description
Yuanqing Yu, Yifan Wang, Weizhi Ma, and colleagues fix a mismatch between how generative recommenders train and how they answer. Their 2026 framework, APAO, targets systems that spell out the next item token by token. Training assumes the true history is always present, yet live search prunes weak early tokens, so the right item can vanish early.
APAO adds training losses at the prefix level, matching how beam search actually discards candidates. An adaptive step focuses effort on whichever prefixes look most at risk during training. Theory supports the design, and tests on several datasets show steady gains across different backbone models. The code is public, so other teams can adopt the fix directly.
Catalog record
| CREATOR | Yuanqing Yu, Yifan Wang, Weizhi Ma, Zhiqiang Guo, Min Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57788 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.02730