LIFT: A Lifecycle-aware Interaction Factorization Transformer for Unified Retrieval and Ranking

2026-09-29

AI Summary

This study addresses the challenge of sharing historical modeling between the retrieval and ranking stages in cascaded recommender systems, which arises from their access to different interaction information. To this end, this work proposes a Lifecycle-Aware Factorized Transformer that decomposes user interactions into ordered state sequences. By leveraging causal sequence modeling, the method enables cross-stage information sharing while preserving stage-specific contexts. Furthermore, it introduces a role-conditioned attention mechanism and lightweight pre-layer normalization biases to unify the historical modeling logic across both stages. Extensive experiments on the ML-20M and Taobao datasets demonstrate that the proposed approach achieves superior joint evaluation scores compared to the strongest baselines, yielding improvements of 4.9% and 3.6%, respectively.

Abstract

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.

Problem

• Recommender Systems
• Retrieval and Ranking
• Cascaded Architecture
• User History Modeling

Innovation

• Lifecycle-aware Interaction Factorization
• Unified Retrieval and Ranking
• Causal Sequence Modeling
• Role-Conditioned Attention
• Pre-LN Bias