Personalized Federated Sequential Recommender

2026-03-22

AI Summary

This work proposes a personalized federated sequential recommendation framework to address the high computational complexity of existing methods, which struggle to balance real-time performance with personalized demands across diverse scenarios. The approach integrates global modeling and local fine-tuning through three key components: an associative Mamba module to enhance sequence modeling efficiency, a variable response mechanism for user-level parameter adaptation, and a dynamic magnitude loss to preserve local personalization signals. By jointly optimizing these elements, the method substantially reduces computational overhead while maintaining recommendation responsiveness and improving personalization accuracy, thereby effectively accommodating heterogeneous user contexts in real-world applications.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Mixture of Experts (MoE)
• Data Mining & Knowledge Management: Recommender Systems

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

In the domain of consumer electronics, personalized sequential recommendation has emerged as a central task. Current methodologies in this field are largely centered on modeling user behavior and have achieved notable performance. Nevertheless, the inherent quadratic computational complexity typical of most existing approaches often leads to inefficiencies that hinder real-time recommendation. Moreover, these methods face challenges in being effectively adapted to the personalized requirements of users across diverse scenarios. To tackle these issues, we propose the Personalized Federated Sequential Recommender (PFSR). In this framework, an Associative Mamba Block is introduced to capture user profiles from a global perspective while improving prediction efficiency. In addition, a Variable Response Mechanism is developed to enable fine-tuning of parameters in accordance with individual user needs. A Dynamic Magnitude Loss is further devised to preserve greater amounts of localized personalized information throughout the training process.

Problem

• personalized sequential recommendation
• computational complexity
• real-time recommendation
• user personalization
• federated learning

Innovation

• Personalized Federated Learning
• Sequential Recommendation
• Associative Mamba Block
• Variable Response Mechanism
• Dynamic Magnitude Loss