面向未来状态控制的推荐世界模型
原标题：Recommendation World Models for Future-State Control
AI 摘要
研究者提出 UA-TWM，一种效用锚定的推荐世界模型接口，用于在序列推荐中评估候选列表动作的未来后果。该方法在冻结的推荐器提供的锚定列表基础上，通过局部后果层估计效用、目标增益和失败风险，并在日志回放与闭环控制两种设置下选择替代列表。在 MovieLens-25M 和 KuaiRand-Pure 上跨十二种序列骨干的匹配比较中，Recall@20、NDCG@20 和未来状态对齐均得到提升，KuaiSim 闭环实验进一步验证了目标导向交互。
正文节选
Recommendation World Models for Future-State Control Abstract Sequential recommendation optimizes which items to rank, while each displayed slate also shapes subsequent feedback and user state. We study how a trained ranker can support decisions about these future consequences. We introduce UA-TWM, a utility-anchored world-model interface that constructs nearby slate actions, estimates their target-relevant consequences, and selects an alternative subject to utility constraints. The reference sl