Towards Efficient Hypergraph and Multi-LLM Agent Recommender Systems

2025-12-06

AI Summary

Generative retrieval in recommendation systems suffers from severe hallucination and high inference overhead. To address these challenges, this paper proposes HGLMRec—a novel framework that pioneers the integration of hypergraph neural networks with multi-LLM agent collaboration to jointly model fine-grained user-item behavioral relations. It further introduces a token-level generative retrieval mechanism to significantly mitigate LLM hallucination. Coupled with a sparsification-based inference strategy, the framework substantially reduces computational cost. Extensive experiments on multiple public benchmarks demonstrate that HGLMRec consistently outperforms state-of-the-art generative recommendation models: it improves Recall@10 by 3.2–7.8%, reduces inference latency by 41.5%, and decreases memory consumption by 36.9%, thereby achieving a balanced optimization of accuracy and efficiency.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recommender Systems (RSs) have become the cornerstone of various applications such as e-commerce and social media platforms. The evolution of RSs is paramount in the digital era, in which personalised user experience is tailored to the user's preferences. Large Language Models (LLMs) have sparked a new paradigm - generative retrieval and recommendation. Despite their potential, generative RS methods face issues such as hallucination, which degrades the recommendation performance, and high computational cost in practical scenarios. To address these issues, we introduce HGLMRec, a novel Multi-LLM agent-based RS that incorporates a hypergraph encoder designed to capture complex, multi-behaviour relationships between users and items. The HGLMRec model retrieves only the relevant tokens during inference, reducing computational overhead while enriching the retrieval context. Experimental results show performance improvement by HGLMRec against state-of-the-art baselines at lower computational cost.

Problem

• Addresses hallucination in generative recommender systems
• Reduces computational cost of multi-LLM agent systems
• Captures complex user-item relationships via hypergraph encoder

Innovation

• Hypergraph encoder captures multi-behavior user-item relationships
• Multi-LLM agent system retrieves only relevant tokens for efficiency
• Reduces computational cost while improving recommendation performance