A Robust and Accessible System for Geospatial Information Extraction using Distil BERT and Fuzzy Logic

This paper proposes a fine-tuned DistilBERT model for accurate and efficient geospatial information extraction from natural language texts. The model is trained on a custom dataset, enhancing its ability to recognize and map colloquial names to canonical forms while differentiating between similar entity names across diverse geographical locations. By leveraging a comprehensive geographical hierarchy analysis and fuzzy logic algorithms, it effectively resolves ambiguity and handles spelling errors and variations in entity names. Integrated with cutting-edge APIs, including Google maps and Wikipedia, the system provides accurate pinning, visualization, and additional information about identified places. Supporting multilingual speech-to-text, it offers users an intuitive interaction experience. Overall, this system represents a notable advancement in geospatial information extraction, offering a valuable reference for further research in location-based natural language processing.