UxSID: Semantic-Aware User Interests Modeling for Ultra-Long Sequence
Researchers have introduced UxSID, a novel framework designed to address the challenging trade-off between efficiency and effectiveness in modeling ultra-long user sequences. Traditional methods typically rely on either item-specific search or item-agnostic compression, but UxSID proposes a third approach utilizing semantic-group shared interest memory. By leveraging Semantic IDs (SIDs) and a dual-level attention strategy, the system captures target-aware preferences without the high computational costs associated with item-specific models. This end-to-end architecture successfully balances computational parsimony with semantic awareness. In large-scale advertising A/B tests, UxSID achieved state-of-the-art performance and demonstrated a tangible business impact with a 0.337% revenue lift. The paper, submitted to arXiv under Computer Science > Artificial Intelligence, highlights significant advancements in user interest modeling for digital advertising and recommendation systems.
Wire timeline
UxSID: Semantic-Aware User Interests Modeling for Ultra-Long Sequence
Researchers have introduced UxSID, a novel framework designed to address the challenging trade-off between efficiency and effectiveness in modeling ultra-long user sequences. Traditional methods typically rely on either item-specific search or item-agnostic compression, but UxSID proposes a third approach utilizing semantic-group shared interest memory. By leveraging Semantic IDs (SIDs) and a dual-level attention strategy, the system captures target-aware preferences without the high computational costs associated with item-specific models. This end-to-end architecture successfully balances computational parsimony with semantic awareness. In large-scale advertising A/B tests, UxSID achieved state-of-the-art performance and demonstrated a tangible business impact with a 0.337% revenue lift. The paper, submitted to arXiv under Computer Science > Artificial Intelligence, highlights significant advancements in user interest modeling for digital advertising and recommendation systems.
cs.AI updates on arXiv.org