突發災害常對社區運作機能及國民生命安全造成重大威脅，建立即時災害地理資訊追蹤成為應對災害的關鍵，能夠協助政府與民眾掌握災害狀況並採取適當應對措施。在資訊通訊科技（Information and Communication Technology, ICT） 蓬勃發展的背景下，社群媒體已成為現代人傳遞訊息的重要管道，並有助於迅速傳播有關災情的訊息。眾多國內外文獻以及實證案例均證實由社群用戶發布的自願者地理資訊（Volunteered Geographic Information, VGI）對災害當下緊急救援與災害管理方面起著重要作用。然而社群文章雜亂無序的資料特性及大量無地理標籤之貼文，導致初步資料處理需要耗費大量時間和人力資源。同時，識別貼文的地理位置亦為災害管理中具有挑戰性的問題。本研究鑒於上述不足之處提出基於BERT的災害辨識與地理定位模型，此模型結合命名實體辨識、文本相似性和空間拓樸關係分析來預測貼文的地理位置。本研究不僅提供了實用的自動化地理定位方法，建立多資訊來源資訊整合方法，並提供空間不確定性範圍，為未來的災害分析與管理提供更強大的支持。
Sudden disasters often pose significant threats to community operations and national public safety. Establishing real-time disaster geographic information tracking is the key to disaster response , as it assists governments and the public in understanding the disaster situation and taking appropriate measures. With the rapid development of information and communication technology (ICT), social media has become a vital channel for modern communication, facilitating the rapid dissemination of information related to disasters. Numerous literature and case studies have confirmed that volunteered geographic information (VGI) released by social media users plays an important role in emergency rescue and disaster management during disasters. However, the chaotic and unstructured nature of social media posts, coupled with many posts lacking geotags, results in significant time and human resource expenditure during initial data processing. At the same time, identifying the location of posts remains a challenging issue in disaster management. In light of these shortcomings, this study proposes a “Disaster Information Geo-tagging Model” based on BERT which combines named entity recognition, relation extraction, and spatial topological analysis to predict the geographic location of posts. This study not only offers practical automated geolocation methods but also accurately represents the real location information described in posts, providing more robust assistance for future disaster analysis and management.
