Smart Map Copilot for Historical Maps with AI-Supported Placename Index by Mapkurator

Keywords: large language models, Mapkurator, historical maps, web service, GeoAI
Abstract Type: Paper Abstract

Abstract

The conversion of scanned maps into a web service is a detailed process that includes scanning, georeferencing, and adding metadata. Traditional map searches relied on exact or fuzzy keyword matches, requiring search terms to be present in the map layer properties. This study leverages large language models like OpenAI to enable more intuitive, free-text queries that interpret both spatial and temporal information. For instance, "Japanese colonial period" can automatically define the years 1895–1945, while "Taipei" is mapped to its current geographic boundaries. A persistent challenge has been the efficient identification of place names across many maps, essential for accurate queries. To address this, we used Mapkurator (PALETTE version) to index archived map tiles. Mapkurator applies over 29,000 place names across 311 maps of Taiwan, including 199 from the Japanese colonial period, alongside 206 annotated maps of 1930s mainland China as references. These place names can be accurately overlaid on the original images using image coordinates. By integrating large language models with Mapkurator, this study introduces a smart spatial search system for historical maps, allowing efficient compilation of map layers on a Web GIS platform and enhancing map search and analysis efficiency.