发表机构
Beijing University of Post and Telecommunications; Southeast University; Xiaohongshu（北京邮电大学; 东南大学; 小红书）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究针对级联推荐系统检索与排序阶段信息异质问题，提出LIFT模型，通过分解交互状态实现两任务共享建模，在ML-20M和Taobao数据集上较基线分别提升4.9%、3.6%联合得分。
AI 中文摘要
级联推荐系统在检索和排序阶段使用相同的用户历史，但两个阶段在交互的不同时点可获取不同信息。我们提出生命周期感知交互分解Transformer（LIFT），将每个交互分解为有序的请求（Request）、物品（Item）、上下文（Context）和动作（Action）状态，并将其建模为因果序列。检索阶段读取请求状态，排序阶段读取上下文状态，使两个任务能共享历史建模，同时保留阶段特定信息。LIFT通过角色条件注意力（Role-Conditioned Attention）和轻量级预层归一化偏置（lightweight Pre-LN Bias）实现该表示。在ML-20M和Taobao数据集上，LIFT在所有评估的联合模型中取得最高联合得分，分别比最强基线提升4.9%和3.6%。损失权重扫描显示出良好的检索-排序权衡，而 ablation 实验和缩放分析进一步研究了生命周期序列构建、模型组件及容量设置。
英文摘要
Cascaded recommender systems use the same user history for retrieval and ranking, while the two stages have access to different information at different points of an interaction. We propose the Lifecycle-aware Interaction Factorization Transformer (LIFT), which decomposes each interaction into ordered Request, Item, Context, and Action states and models them as a causal sequence. Retrieval reads the Request state, while ranking reads the Context state, allowing both tasks to share history modeling while preserving stage-specific information. LIFT instantiates this representation with Role-Conditioned Attention and a lightweight Pre-LN Bias. On ML-20M and Taobao, LIFT achieves the highest Joint Score among the evaluated joint models, improving over the strongest baselines by 4.9% and 3.6%, respectively. Loss-weight sweeps show favorable retrieval--ranking trade-offs, while ablations and scaling analyses further examine lifecycle sequence construction, model components, and capacity settings.
Comments16 pages, 5 figures, 7 tables
