历史误导下的非对称边际监督：指令引导LLM生成式推荐
原标题：When History Misleads: Asymmetric Margin Supervision for Instruction-Guided LLM Generative Recommendation
AI 摘要
该论文针对指令引导的LLM生成式推荐中历史事件与用户当前请求冲突的问题，提出Asymmetric Intervention-Guided Margin Supervision (AIMS)方法。AIMS利用冻结参考模型评估单条历史事件删除对目标项得分及其与边界竞争项差距的影响，将满足条件的删除所对应的最大边际作为请求特定的反事实监督信号，训练时保留完整历史作为输入，并采用非对称竞争抑制损失。在工业数据集和两个公开基准上，AIMS在六个LLM骨干上提升了Recall和NDCG，消融实验支持请求特定边际与非对称监督的有效性。
正文节选
When History Misleads: Asymmetric Margin Supervision for Instruction-Guided LLM Generative Recommendation Abstract In instruction-guided generative recommendation, LLM-based recommenders need to balance two goals: responding to the user’s current request and aligning with the preferences in their interaction history. When the two conflict, history events can override the request. We show that turning the effect of individual history events into supervision faces two obstacles. First, the events