Using natural language processing technologies to automate the coordination of objects from travelogs

The authors discuss the concept of travelogue and its connection with spatial data, the importance of using this type of artistic work, including creation of cartographic materials. In addition to manual processing of travelogue texts, algorithms and natural language methods can be used for automation. Existing examples of manipulations containing geographical names and integrating the results into databases of GIS are considered. The degree of automation using natural language processing (NLP) technologies and the accuracy of creating a route database for a random set of documents is also analyzed. Expanding the parameters of the Movement Action Events object and a quantitative assessment of their relevance among the elements with similar values are proposed. The analyzed set of documents contains 150 texts written between the 18th and 20th centuries, with a total volume of 46 thousand words (with average number of 308 per document including addresses). The results of comparing several pre-trained neural networks variants (large language models as well) are presented in terms of the final accuracy of determining travel routes using the given texts, including those additionally trained using an additional data set