Agent4POI: Agentic Context-Conditioned Affordance Reasoning for Multimodal Point-of-Interest Recommendation
The article introduces Agent4POI, a novel framework for point-of-interest (POI) recommendation that generates context-sensitive multimodal representations. Unlike traditional systems that rely on static embeddings, Agent4POI dynamically creates representations based on situational context. The framework demonstrates significant performance improvements over existing methods in various evaluation scenarios.
- ▪Agent4POI is the first POI recommendation framework that generates context-conditioned multimodal representations at recommendation time.
- ▪The framework achieves a 23.2% relative gain over the strongest baseline on three POI benchmarks.
- ▪In cold-start scenarios, Agent4POI outperforms the best content-based baseline by up to 2.4 times.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Information Retrieval arXiv:2605.15203 (cs) [Submitted on 3 Apr 2026] Title:Agent4POI: Agentic Context-Conditioned Affordance Reasoning for Multimodal Point-of-Interest Recommendation Authors:Jinze Wang, Yangchen Zeng, Tiehua Zhang, Lu Zhang, Yuze Liu, Yongchao Liu, Xingjun Ma, Zhu Sun View a PDF of the paper titled Agent4POI: Agentic Context-Conditioned Affordance Reasoning for Multimodal Point-of-Interest Recommendation, by Jinze Wang and 7 other authors View PDF HTML (experimental) Abstract:We introduce Agent4POI, the first POI recommendation framework that generates context-conditioned multimodal representations at recommendation time, rather than relying on static POI embeddings pre-computed independently of context.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
