← 研踪 PaperTrailarXiv:2611.15445 · v1
ReAlign: Reward-Guided Preference Alignment for Generative Re-Ranking
首次提交 2026-11-11 21:16
版本更新 2026-11-11 21:16
雷达观测 2026-11-13 11:11
摘要
Re-ranking refines the upstream ranked list to maximize listwise user utility. Existing two-stage generator-evaluator methods decouple list generation from list evaluation, leaving the generator without direct access to the listwise utility signal, while scoring-based one-stage methods score items under a fixed upstream permutation context without explicitly searching the permutation space. We propose \textbf{ReAlign}, a reward-guided preference alignment framework that couples a listwise reward model with a candidate-aware autoregressive generator. The reward model decomposes the listwise reward into position-wise contributions, modeling user exposure through a survival-based exposure module and item value through per-objective heads for heterogeneous engagement signals. The generator is trained in two stages: supervised pre-training on logged exposure lists to establish a relevance prior, followed by reward-guided post-training that aligns the generator with the reward model through complementary distribution and ranking alignment over within-pool standardized rewards. Extensive offline experiments show that ReAlign consistently outperforms existing state-of-the-art baselines.
作者备注：9 pages, 3 figures, ACM conference format
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-11 21:16 · 最新观测版本
ReAlign: Reward-Guided Preference Alignment for Generative Re-Ranking
Re-ranking refines the upstream ranked list to maximize listwise user utility. Existing two-stage generator-evaluator methods decouple list generation from list evaluation, leaving the generator without direct access to the listwise utility signal, while scoring-based one-stage methods score items under a fixed upstream permutation context without explicitly searching the permutation space. We propose \textbf{ReAlign}, a reward-guided preference alignment framework that couples a listwise reward model with a candidate-aware autoregressive generator. The reward model decomposes the listwise reward into position-wise contributions, modeling user exposure through a survival-based exposure module and item value through per-objective heads for heterogeneous engagement signals. The generator is trained in two stages: supervised pre-training on logged exposure lists to establish a relevance prior, followed by reward-guided post-training that aligns the generator with the reward model through complementary distribution and ranking alignment over within-pool standardized rewards. Extensive offline experiments show that ReAlign consistently outperforms existing state-of-the-art baselines.
查看此版本解读 →查看此版本原文 ↗