RecSoup：面向可控多目标推荐的高效模型合并方法
原标题：RecSoup: Efficient Model Merging Approach for Controllable Multi-Objective Recommendation
AI 摘要
该论文提出 RecSoup，一种用于可控多目标推荐（CMOR）的轻量级两阶段模型合并方法。它先通过同时考虑目标特定重要性和跨目标相关性的剪枝策略缓解参数干扰，再通过可学习的偏好到系数映射实现自适应参数合并，从而在推理时动态控制目标权衡，无需联合重训练或额外生成模块。理论上证明其能完整遍历帕累托前沿，实验显示帕累托前沿质量与可控性提升，推理延迟比参数生成方法快 165–228 倍，代码已开源。
正文节选
RecSoup: Efficient Model Merging Approach for Controllable Multi-Objective Recommendation Abstract. Recommender systems often need to balance multiple conflicting objectives. Controllable Multi-Objective Recommendation (CMOR) enables dynamic adjustment of objective trade-offs at inference time without retraining, adapting to shifting user or system requirements. Existing CMOR approaches typically rely on output-level reranking or parameter-generating modules, which either limit the scope of fea