UxSID: Semantic-Aware User Interests Modeling for Ultra-Long Sequence

2026-05-09

AI Summary

This work addresses the challenge of balancing efficiency and effectiveness in modeling ultra-long user behavior sequences by proposing UxSID, a novel end-to-end interest memory architecture. UxSID introduces semantic IDs (SIDs) and a two-level attention mechanism to enable semantic-group-based sharing of interest memory pathways, thereby overcoming the limitations of traditional item-by-item retrieval or indiscriminate sequence compression. This design facilitates fine-grained, target-aware user preference modeling while substantially reducing computational overhead. Evaluated in large-scale online A/B tests on a real-world advertising system, UxSID achieves state-of-the-art performance and delivers a 0.337% increase in ad revenue.

Abstract

Modeling ultra-long user sequences involves a difficult trade-off between efficiency and effectiveness. While current paradigms rely on either item-specific search or item-agnostic compression, we propose UxSID, a framework exploring a third path: semantic-group shared interest memory. By utilizing Semantic IDs (SIDs) and a dual-level attention strategy, UxSID captures target-aware preferences without the heavy cost of item-specific models. This end-to-end architecture balances computational parsimony with semantic awareness, achieving state-of-the-art performance and a 0.337% revenue lift in large-scale advertising A/B test.

Problem

• ultra-long sequence
• user interests modeling
• efficiency-effectiveness trade-off
• semantic awareness
• computational efficiency

Innovation

• Semantic ID
• shared interest memory
• ultra-long sequence modeling
• dual-level attention
• user interest modeling