arXiv cs.CL· Sunkyung Lee, Xiaokai Wei, Da Long, Hyunsik Jeon, Jiajun Wu, Rithwik Seth, Jongwuk Lee, Michelle Gong, Gungor Polatkan·· 11 小时前AI 评分29
SteerGR：用稀疏自编码器实现可控生成式游戏推荐
SteerGR: Steering Generative Recommendation via Sparse Autoencoders for Controllable Game Discovery
AI 导读
SteerGR 框架通过稀疏自编码器实现生成式推荐的概念级控制，可在不重新训练推荐模型的情况下放大或抑制选定特征，使推荐结果靠近或远离目标概念。该框架分三阶段：用 LLM 定义并锚定游玩概念、按概念关联发现并依据引导效果筛选特征、对稀疏特征进行干预。在两个工业级生成式推荐模型和真实游戏数据集上的实验显示，所选特征能有效引导推荐方向。
正文 · 原文
Abstract:Generative recommendation aims to predict item identifiers from user interaction histories, but explicit control over the properties of recommended items remains underexplored. A key challenge is to connect human-readable concepts to internal features whose interventions effectively change the recommended items. To address this, we propose Steerable Generative Recommender (SteerGR), a framework that uses sparse autoencoders for concept-level control in game discovery. SteerGR proceeds through three stages: (i) Concept Definition and Grounding for defining play concepts with an LLM; (ii) Feature Discovery and Selection for discovering candidate features by their concept associations and selecting one by its steering effect; and (iii) Sparse Feature Intervention for amplifying or suppressing selected features to steer recommendations without retraining the recommender. Extensive experiments with two industrial generative recommendation models on a real-world game dataset show that the selected features successfully steer recommendations toward or away from target concepts. We believe this work takes a first step to extend SAE-based steering from LLMs to industrial generative recommendation.
| Comments: | Preprint (15 pages, 11 figures, 11 tables) |
| Subjects: | Information Retrieval (cs.IR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.08790 [cs.IR] |
| (or arXiv:2611.08790v1 [cs.IR] for this version) | |
| https://doi.org/10.48550/arXiv.2611.08790 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Sunkyung Lee [view email]
[v1]
Thu, 5 Nov 2026 23:25:08 UTC (2,564 KB)
来源：arXiv cs.CL · arxiv.org