Elamur Research⌕
☰
home / papers / 2026 / Similar Region Search using LLMs on Spatial Feature Space

Similar Region Search using LLMs on Spatial Feature Space

Paper · 2026 · Al-Amin Sany, Mohaiminul Islam, Tanzima Hashem et al.

Summary

The paper tackles similar region search (SRS): given a query region, rank candidate urban regions by how closely they resemble it. The authors argue that distance formulas, region embeddings and deep metric learning reduce regions to bare numbers. These methods miss functional equivalence, such as two residential areas with different store layouts, and they struggle to merge heterogeneous data. Their framework, LLM4SRS, turns each region's spatial attributes into a textual description and asks an LLM (LLaMA-3) to output a similarity ranking directly. The attributes are POI distributions, check-in activity and building footprints. No separate ranking algorithm is needed. To avoid relying on labeled data, they perturb an anchor region's features with controlled noise. This creates candidates of known similarity, which are then used for self-supervised fine-tuning. On real city data from Singapore and Tokyo, including cross-city tests, LLM4SRS beats embedding and metric-learning baselines on HR@k and MRR@k. The work suggests LLMs can reason contextually over mixed spatial data, though inference costs more than lightweight baselines.

Key points

  • Inputs combine three feature types: POI attributes, check-in data and building footprints. In the ablation, using all three works best, dropping building footprints hurts more than dropping check-ins, and POI alone performs worst.
  • Evaluation varies noise ratios from 0.1 to 0.5, uses 10 or 20 candidates and k = 1, 3, 5, 7. Scores are slightly lower with 20 candidates, but LLM4SRS stays ahead of the baselines.
  • Reported gains include about 20% higher MRR@3 than Triplet (Singapore, 10 candidates) and 18% higher HR@7 than CARE (Tokyo, 20 candidates). At k = 1, it is roughly 15% better than Sentence Transformer on average.
  • The backbone is LLaMA-3, with quantization and parameter-efficient fine-tuning. Code is released at github.com/LLM4SRS.
  • The authors list several limitations:
    • Ground truth comes from synthetic perturbations, with no human labels.
    • Only one backbone LLM was tested.
    • Results may be sensitive to prompt wording.
    • Imagery and road networks are not used.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Bangladesh University of Engineering and TechnologyDhaka · Bangladesh7 works · 308 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksSignal Processing · 493 worksTransportation · 7 works

Cite

Cite as Al-Amin Sany, Mohaiminul Islam, Tanzima Hashem, Md. Ashraful Islam and Mohammed Eunus Ali. 2026. Similar Region Search using LLMs on Spatial Feature Space. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.201

@inproceedings{sany2026similar,
  title = {{Similar Region Search using LLMs on Spatial Feature Space}},
  author = {Al-Amin Sany, Mohaiminul Islam, Tanzima Hashem, Md. Ashraful Islam and Mohammed Eunus Ali},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.201},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.201}
}

References

31 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Deniz Canturk, Pinar Karagoz, Sang-Wook Kim, and Ismail Hakki Toroslu. 2023. Trust-aware location recommendation in location-based social networks: A graph-based approach. Expert Systems with Applications, 213:119048. find
  2. Gang Cao, Shengmin Cui, and Inwhee Joe. 2023. Improving the spatial–temporal aware attention network with dynamic trajectory graph learning for next point-of-interest recommendation. Information Processing & Management, 60(3):103335. find
  3. Weiliang Chan and Qianqian Ren. 2023. Region-wise attentive multi-view representation learning for urban region embedding. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, pages 3763–3767. find
  4. Michael Han Daniel Han and Unsloth team. 2023. Unsloth. find
  5. Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, and Christopher Ré. 2022. Flashattention: Fast and memory-efficient exact attention with io-awareness. Preprint, arXiv:2205.14135. find
  6. Kaiyu Feng, Gao Cong, Christian S Jensen, and Tao Guo. 2019. Finding attribute-aware similar region for data analysis. Proceedings of the VLDB Endowment, 12(11):1414–1426. find
  7. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. Preprint, arXiv:2106.09685.
  8. Bo Hui, Da Yan, Wei-Shinn Ku, and Wenlu Wang. 2020. Predicting economic growth by region embedding: A multigraph convolutional network approach. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management, pages 555–564. find
  9. Benoit Jacob, Skirmantas Kligys, Bo Chen, Menglong Zhu, Matthew Tang, Andrew Howard, Hartwig Adam, and Dmitry Kalenichenko. 2017. Quantization and training of neural networks for efficient integer-arithmetic-only inference. Preprint, arXiv:1712.05877. find
  10. Yue Jiang, Qin Chao, Yile Chen, Xiucheng Li, Shuai Liu, and Gao Cong. 2024. Urbanllm: Autonomous urban activity planning and management with large language models. arXiv preprint arXiv:2406.12360. find
  11. Jiahui Jin, Yifan Song, Dong Kan, Binjie Zhang, Yan Lyu, Jinghui Zhang, and Hongru Lu. 2024a. Learning context-aware region similarity with effective spatial normalization over point-of-interest data. Information Processing & Management, 61(3):103673. find
  12. Jiahui Jin, Yifan Song, Dong Kan, Binjie Zhang, Yan Lyu, Jinghui Zhang, and Hongru Lu. 2024b. Learning context-aware region similarity with effective spatial normalization over point-of-interest data. Information Processing & Management, 61(3):103673. find
  13. Xiongnan Jin, Byungkook Oh, Sanghak Lee, Dongho Lee, Kyong-Ho Lee, and Liang Chen. 2019. Learning region similarity over spatial knowledge graphs with hierarchical types and semantic relations. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management, pages 669… find
  14. Bohan Li, Hao Zhou, Junxian He, Mingxuan Wang, Yiming Yang, and Lei Li. 2020. On the sentence embeddings from pre-trained language models. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9119–9130, Online. Association for Computational Linguis…
  15. Peibo Li, Maarten de Rijke, Hao Xue, Shuang Ao, Yang Song, and Flora D Salim. 2024a. Large language models for next point-of-interest recommendation. In Proceedings of the 47th International ACM SI-GIR Conference on Research and Development in Information Retrieval, pages 1463–1472. find
  16. Yi Li, Weiming Huang, Gao Cong, Hao Wang, and Zheng Wang. 2023. Urban region representation learning with openstreetmap building footprints. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 1363–1373. find
  17. Zechen Li, Weiming Huang, Kai Zhao, Min Yang, Yongshun Gong, and Meng Chen. 2024b. Urban region embedding via multi-view contrastive prediction. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 8724–8732. find
  18. Zhonghang Li, Lianghao Xia, Jiabin Tang, Yong Xu, Lei Shi, Long Xia, Dawei Yin, and Chao Huang. 2024c. Urbangpt: Spatio-temporal large language models. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 5351–5362. find
  19. Yiding Liu, Kaiqi Zhao, and Gao Cong. 2018. Efficient similar region search with deep metric learning. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 1850–1859. find
  20. Yu Liu, Jingtao Ding, Yanjie Fu, and Yong Li. 2023. Urbankg: An urban knowledge graph system. ACM Transactions on Intelligent Systems and Technology, 14(4):1–25. fine-tuning methods for pretrained language models: A critical review and assessment. Preprint, arXiv:2312.12148. find
  21. Liang Zhang, Cheng Long, and Gao Cong. 2022. Region embedding with intra and inter-view contrastive learning. IEEE Transactions on Knowledge and Data Engineering, 35(9):9031–9036. find
  22. Siyao Zhang, Daocheng Fu, Wenzhe Liang, Zhao Zhang, Bin Yu, Pinlong Cai, and Baozhen Yao. 2024. Trafficgpt: Viewing, processing and interacting with traffic foundation models. Transport Policy, 150:95– 105. find
  23. Yifan Zhang, Cheng Wei, Shangyou Wu, Zhengting He, and Wenhao Yu. 2023. Geogpt: understanding and processing geospatial tasks through an autonomous gpt. arXiv preprint arXiv:2307.07930. — [arXiv:2307.07930](https://arxiv.org/abs/2307.07930) find
  24. Silin Zhou, Dan He, Lisi Chen, Shuo Shang, and Peng Han. 2023. Heterogeneous region embedding with prompt learning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 4981–4989. find
  25. Rohin Manvi, Samar Khanna, Gengchen Mai, Marshall Burke, David Lobell, and Stefano Ermon. 2023. Geollm: Extracting geospatial knowledge from large language models. arXiv preprint arXiv:2310.06213. find
  26. Grant McKenzie and Daniel Romm. 2021. Measuring urban regional similarity through mobility signatures. Computers, Environment and Urban Systems, 89:101684. find
  27. Chang Sheng, Yu Zheng, Wynne Hsu, Mong Li Lee, and Xing Xie. 2010. Answering top-k similar region queries. In Database Systems for Advanced Applications: 15th International Conference, DASFAA 2010, Tsukuba, Japan, April 1-4, 2010, Proceedings, Part I 15, pages 186–201. Springer. find
  28. Fengze Sun, Jianzhong Qi, Yanchuan Chang, Xiaoliang Fan, Shanika Karunasekera, and Egemen Tanin. 2024. Urban region representation learning with attentive fusion. In 2024 IEEE 40th International Conference on Data Engineering (ICDE), pages 4409–4421. IEEE. find
  29. Xigang Sun, Jingya Zhou, Ling Liu, and Wenqi Wei. 2023. Explicit time embedding based cascade attention network for information popularity prediction. Information Processing & Management, 60(3):103278. find
  30. Shangbin Wu, Xu Yan, Xiaoliang Fan, Shirui Pan, Shichao Zhu, Chuanpan Zheng, Ming Cheng, and Cheng Wang. 2022. Multi-graph fusion networks for urban region embedding. arXiv preprint arXiv:2201.09760. — [arXiv:2201.09760](https://arxiv.org/abs/2201.09760) find
  31. Lingling Xu, Haoran Xie, Si-Zhao Joe Qin, Xiaohui Tao, and Fu Lee Wang. 2023. Parameter-efficient *Figure 7: Prompt construction pipeline.* find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sany-2026-similar-region-search-llms-spatial/