Extended Abstract: Entity Linking in GeoKBQA: Deep Learning Approaches for Enhanced Mention Detection

This study addresses the integration of entity linking into geographic information-based question answering (GeoKBQA) systems, enhancing their performance using deep learning models for mention detection (MD). BERT, RoBERTa, and SpanBERT were fine-tuned on the NLMAPS dataset, which contains geographical questions for MD tasks, while ChatGPT utilized the 'few-shot prompting' approach. Fine-tuned BERT, RoBERTa, and SpanBERT models achieved Fl scores of 0.96, 0.97, and 0.89, while ChatGPT scored 0.99, demonstrating its effectiveness.