A Multimodal Approach to Geolocation Extraction: Leveraging Wikipedia, Person-Specific Data, and NLP for Enhanced Geographic Entity Recognition

This paper presents a multimodal approach to geolocation extraction that integrates Wikipedia content, person-specific data, and NLP techniques for enhanced geographic entity recognition. Accurate geolocation of person entities is vital in digital humanities, knowledge graph enrichment, and geographic information systems (GIS). In this study, we investigate in a novel way, how adding names to geotagging improves engagement and personalization. The proposed framework retrieves Wikipedia pages of individuals, applies named entity recognition (NER) to extract location mentions, and maps them to precise geographic coordinates using geocoding (Nominatim) with knowledge graph-based disambiguation. Evaluation on a curated dataset of individuals demonstrates the effectiveness of this approach, achieving a precision of 1.00, recall of 0.94, F1-score of 0.97, and overall accuracy of 0.94. This method is helpful for applications in historical research, the social sciences, and location-based data analysis by dynamically extracting and mapping significant locations associated with individuals. These results highlight the potential of combining structured and unstructured resources with NLP to enhance the accuracy and robustness of geotagging across diverse application domains.