HyperAgent4POI: Dynamic Semantic Message Passing on Multi-Agent Hypergraphs for Missing-Modality Recommendation

2026-08-03

AI Summary

This work addresses the challenge of weakened semantic representation and unreliable high-order interaction modeling in point-of-interest (POI) recommendation under real-world scenarios where textual or visual modalities are often missing. To this end, we propose a multi-agent hypergraph framework that uniquely integrates a frozen LLaMA large language model with hypergraph neural networks. Our approach employs role-specific adapters to generate semantic messages from nodes to hyperedges and introduces a dynamic semantic message-passing mechanism that simultaneously achieves cross-modal completion and soft relational refinement within each hypergraph layer. A representation caching strategy further enables efficient online inference without requiring real-time LLM calls. Experiments on three real-world location-based social network datasets demonstrate that our method consistently outperforms 15 strong baselines, achieving an average 8.2% improvement in NDCG@20 even under a 60% modality missing rate.

Abstract

Next Point-of-Interest (POI) recommendation benefits from textual and visual content that describes venue semantics, yet such content is often incomplete in real-world services. Missing modalities weaken POI representations and reduce the semantic evidence available for ranking. The resulting representations also provide unreliable evidence for modeling higher-order user--POI interactions. We propose HyperAgent4POI, which uses Dynamic Semantic Message Passing (DSMP) to perform modality completion and soft incidence refinement within each hypergraph layer. Persistent node agents share a frozen Llama backbone and use role-specific adapters to produce node-to-hyperedge messages. Semantic hyperedge motifs formed from these messages guide soft incidence scoring and modality completion. Final node representations are cached for online ranking without LLM calls. Experiments on three real-world LBSN datasets show consistent ranking gains over 15 baselines across modality-missing rates, while cached inference provides practical online efficiency. Under a 60% modality-missing rate, HyperAgent4POI improves NDCG@20 over the strongest baseline by 8.2% on average across the three datasets.

Problem

• missing-modality recommendation
• Point-of-Interest recommendation
• semantic representation
• multi-modal data incompleteness
• user-POI interaction

Innovation

• Dynamic Semantic Message Passing
• Multi-Agent Hypergraphs
• Missing-Modality Recommendation
• Soft Incidence Refinement
• Cached Inference