Scholarlink

<p>The increasing availability of geospatial data from social media has created new opportunities for location-based analysis. However, the specific characteristics of such data limits traditional analytical methods. This study reviews recent efforts to integrate large language models (LLMs) with geospatial social media analytics, synthesizing 20 studies published in 2024 and 2025. The review examines application domains, modeling approaches, data characteristics, and integration strategies. Findings show that LLMs enhance analysis in disaster management, urban planning, and the study of semantic footprints and human behavior by enabling tasks such as location inference from unstructured text and extending analysis beyond explicitly geotagged posts. GPT-based models dominate current research (used in 10 studies), with X (formerly Twitter) as the primary data source in 10 studies. Dataset scales range from approximately 4,000 to 210&#xa0;million records, with study areas spanning localized urban districts to multi-country scales. Key limitations include spatial ambiguity, geographic and linguistic biases, limited interpretability, high computational costs, and unresolved privacy and ethical concerns. Overall, this review highlights both the promise of LLMs for geospatial social media analytics and the critical limitations that must be addressed to support robust and responsible applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Emerging uses of large language models for geospatial social media analytics: a review

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • Farbod Farhangi,
  • Ali Asghar Alesheikh,
  • Abolghasem Sadeghi-Niaraki,
  • Seyed Vahid Razavi-Termeh,
  • Biswajeet Pradhan

摘要

The increasing availability of geospatial data from social media has created new opportunities for location-based analysis. However, the specific characteristics of such data limits traditional analytical methods. This study reviews recent efforts to integrate large language models (LLMs) with geospatial social media analytics, synthesizing 20 studies published in 2024 and 2025. The review examines application domains, modeling approaches, data characteristics, and integration strategies. Findings show that LLMs enhance analysis in disaster management, urban planning, and the study of semantic footprints and human behavior by enabling tasks such as location inference from unstructured text and extending analysis beyond explicitly geotagged posts. GPT-based models dominate current research (used in 10 studies), with X (formerly Twitter) as the primary data source in 10 studies. Dataset scales range from approximately 4,000 to 210 million records, with study areas spanning localized urban districts to multi-country scales. Key limitations include spatial ambiguity, geographic and linguistic biases, limited interpretability, high computational costs, and unresolved privacy and ethical concerns. Overall, this review highlights both the promise of LLMs for geospatial social media analytics and the critical limitations that must be addressed to support robust and responsible applications.