POI-Enhancer: An LLM-based Semantic Enhancement Framework for POI Representation Learning
Jiawei Cheng, Jingyuan Wang, Yichuan Zhang, Jiahao Ji, Yuanshao Zhu, Zhibo Zhang, Xiangyu Zhao
TL;DR
This work presents POI-Enhancer, an LLM-based framework that enriches POI representations by extracting semantic knowledge from frozen LLMs through three targeted prompts and integrating it via a dual-feature alignment, semantic fusion, and cross-attention fusion pipeline. A multi-view contrastive learning scheme with Sequence-Time, Geography, and Functional sampling further injects human-understandable semantics while preserving original structure. Extensive experiments on three public check-in datasets show consistent improvements over strong baselines across POI recommendation, check-in sequence classification, and visitor flow prediction, with ablations confirming the contributions of prompts, alignment, fusion, and sampling strategies. The approach enables richer POI embeddings that better capture spatial-temporal patterns and semantic context, offering practical benefits for mobility analytics and smart-city applications.
Abstract
POI representation learning plays a crucial role in handling tasks related to user mobility data. Recent studies have shown that enriching POI representations with multimodal information can significantly enhance their task performance. Previously, the textual information incorporated into POI representations typically involved only POI categories or check-in content, leading to relatively weak textual features in existing methods. In contrast, large language models (LLMs) trained on extensive text data have been found to possess rich textual knowledge. However leveraging such knowledge to enhance POI representation learning presents two key challenges: first, how to extract POI-related knowledge from LLMs effectively, and second, how to integrate the extracted information to enhance POI representations. To address these challenges, we propose POI-Enhancer, a portable framework that leverages LLMs to improve POI representations produced by classic POI learning models. We first design three specialized prompts to extract semantic information from LLMs efficiently. Then, the Dual Feature Alignment module enhances the quality of the extracted information, while the Semantic Feature Fusion module preserves its integrity. The Cross Attention Fusion module then fully adaptively integrates such high-quality information into POI representations and Multi-View Contrastive Learning further injects human-understandable semantic information into these representations. Extensive experiments on three real-world datasets demonstrate the effectiveness of our framework, showing significant improvements across all baseline representations.