LASER: An Efficient Target-Aware Segmented Attention Framework for End-to-End Long Sequence Modeling

2026-02-12

AI Summary

This work addresses the challenges of high I/O latency and the quadratic complexity of attention mechanisms in modeling ultra-long user behavior sequences by proposing LASER, a framework that achieves efficient end-to-end learning through co-optimization of system and algorithmic design. Its key innovations include SeqVault, a unified storage architecture leveraging a hybrid DRAM-SSD index to substantially reduce I/O overhead, and Segmented Target Attention (STA), which integrates Sigmoid gating with a lightweight global Stacked Target Attention (GSTA) to compress sequences while preserving critical interest signals. Experimental results demonstrate that LASER outperforms existing state-of-the-art methods on offline metrics, and large-scale online A/B tests involving over 100 million daily active users show a 2.36% increase in ADV V and a 2.08% boost in revenue.

Technology Category

• Search and Optimization: Learning to Search
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Online Learning & Bandits

Application Category

• User Modeling, Personalization and Recommendation: Studies of user behavior, including longitudinal effects of personalized systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Modeling ultra-long user behavior sequences is pivotal for capturing evolving and lifelong interests in modern recommendation systems. However, deploying such models in real-time industrial environments faces a strict"Latency Wall", constrained by two distinct bottlenecks: the high I/O latency of retrieving massive user histories and the quadratic computational complexity of standard attention mechanisms. To break these bottlenecks, we present LASER, a full-stack optimization framework developed and deployed at Xiaohongshu (RedNote). Our approach tackles the challenges through two complementary innovations: (1) System efficiency: We introduce SeqVault, a unified schema-aware serving infrastructure for long user histories. By implementing a hybrid DRAM-SSD indexing strategy, SeqVault reduces retrieval latency by 50% and CPU usage by 75%, ensuring millisecond-level access to full real-time and life-cycle user histories. (2) Algorithmic efficiency: We propose a Segmented Target Attention (STA) mechanism to address the computational overhead. Motivated by the inherent sparsity of user interests, STA employs a sigmoid-based gating strategy that acts as a silence mechanism to filter out noisy items. Subsequently, a lightweight Global Stacked Target Attention (GSTA) module refines these compressed segments to capture cross-segment dependencies without incurring high computational costs. This design performs effective sequence compression, reducing the complexity of long-sequence modeling while preserving critical signals. Extensive offline evaluations demonstrate that LASER consistently outperforms state-of-the-art baselines. In large-scale online A/B testing serving over 100 million daily active users, LASER achieved a 2.36% lift in ADVV and a 2.08% lift in revenue, demonstrating its scalability and significant commercial impact.

Problem

• long sequence modeling
• recommendation systems
• latency bottleneck
• attention mechanism
• user behavior sequences

Innovation

• Segmented Target Attention
• SeqVault
• Long Sequence Modeling
• Hybrid DRAM-SSD Indexing
• Target-Aware Attention