从物品分数到语义Token：生成式推荐的测试时协同适配
原标题：From Item Scores to Semantic Tokens: Test-Time Collaborative Adaptation for Generative Recommendation
AI 摘要
论文提出 TCBSA（Test-Time Collaborative Beam Search Adaptation），一种无需训练的方法，将独立训练的判别式推荐器给出的物品级协同分数，转化为语义前缀共享边上的势能，并通过置信度门控在 beam 选择前调节其对路径累积分数的贡献。该方法旨在解决生成式推荐中 beam search 过早剪枝新兴兴趣路径的问题，在冻结生成器与判别器的情况下实现测试时协同适配。全目录实验显示，该方法在多个数据集及不同生成式/判别式推荐器组合上均优于强基线，并改善了目标路径存活率与推理延迟的平衡。
正文节选
From Item Scores to Semantic Tokens: Test-Time Collaborative Adaptation for Generative Recommendation Abstract. Generative recommenders using semantic identifiers predict items through successive token decisions, but beam search may irreversibly prune promising item paths. This limitation is particularly consequential when a few recent interactions signal an emerging interest while earlier behavior favors established preferences. Yet an independently trained discriminative recommender may rank t