POI Recommendation with LLM-Augmented Multi-Graph Learning and Contrastive Alignment
2026-08-17 • Information Retrieval
Information RetrievalMachine Learning
AI summaryⓘ
The authors developed a recommendation system called LLM-MGCL that helps suggest places to visit even if those places don't have many reviews or user interactions. They do this by combining information from three different graphs: one based on user behavior, one on the meaning of item descriptions generated by a language model, and one on the physical locations of places. Their method connects and learns from these different views to better understand items with little data. Tests on Yelp data show their model outperforms older methods and helps solve the problem of recommending new or less-known places.
Point-of-Interest (POI) RecommendationGraph Neural Networks (GNN)Cold-start ProblemLightGCNContrastive LearningInfoNCE ObjectiveSemantic GraphGeographic GraphSentence EmbeddingsHaversine Distance
Authors
Burak Tamer, Wolfram Höpken, Zehui Wang
Abstract
Point-of-interest (POI) recommendation models based on graph neural networks achieve strong performance by propagating collaborative signals over user-item interactions, yet they struggle with the cold-start problem, where items with few or no interactions are not represented. In this paper, we propose LLM-augmented Multi-Graph Contrastive Learning (LLM-MGCL), a multi-graph neural network that uses semantic and spatial information about items to extend the LightGCN backbone with two auxiliary item-item graphs: a semantic graph constructed from sentence embeddings of LLM-generated photo summaries and keywords, and a geographic graph derived from Haversine distances between business locations. Item embeddings are propagated over all three graphs in parallel, fused additively, and aligned across views through a bidirectional InfoNCE contrastive objective that connects behavioral, semantic, and spatial representations of the same items. Experiments on the Yelp Multimodal Recommendation Dataset show that LLM-MGCL outperforms classical collaborative filtering, matrix factorization, and interaction-only graph neural network baselines. It improves Recall@20 by 52.0% and NDCG@20 by 64.8% over LightGCN while performing on par with the strongest contrastive baseline, Self-supervised Graph Learning (SGL), which is also affected by the cold-start problem. An ablation study reveals that the cross-view contrastive alignment (CA) is the primary driver of these gains, with the best performance achieved when all three graphs are combined. Our results suggest that externally grounded, LLM-derived item knowledge can effectively compensate for missing collaborative signal and mitigate the item cold-start problem in POI recommendation.