UxSID: Semantic-Aware User Interest Modeling for Ultra-Long Sequences
A new framework called UxSID (Semantic-Aware User Interests Modeling for Ultra-Long Sequence) has been proposed to address the trade-off between efficiency and effectiveness in modeling ultra-long user sequences. Unlike existing paradigms that rely on item-specific search or item-agnostic compression, UxSID introduces a third approach: semantic-group shared interest memory. It uses Semantic IDs (SIDs) and a dual-level attention strategy to capture target-aware preferences without the heavy computational cost of item-specific models. The end-to-end architecture balances computational parsimony with semantic awareness. In a large-scale advertising A/B test, UxSID achieved state-of-the-art performance and a 0.337% revenue lift. The paper is available on arXiv under the Computer Science > Artificial Intelligence category.
Key facts
- UxSID stands for Semantic-Aware User Interests Modeling for Ultra-Long Sequence.
- The framework uses Semantic IDs (SIDs) and a dual-level attention strategy.
- It achieves a 0.337% revenue lift in a large-scale advertising A/B test.
- The paper is categorized under Computer Science > Artificial Intelligence on arXiv.
- UxSID explores a third path beyond item-specific search and item-agnostic compression.
- The architecture is end-to-end and balances computational parsimony with semantic awareness.
- It captures target-aware preferences without heavy computational cost.
- The framework achieves state-of-the-art performance.
Entities
Institutions
- arXiv