Automation of data preparation for mapping using natural language processing systems

Abstract

The current level of development of information technology makes it possible to automate the processing of those types of data that only a specialist could previously work with. One such example is natural language processing technologies that implement the functions of sentiment analysis, machine translation, and question-answer systems. For the processes of creating cartographic and geoinformation works, the methods of extracting named entities are of the greatest interest, which allows extracting geographical names from unstructured text and linking named entities, which make it possible to create logical links between the extracted names of spatial objects. Their processing, through a local or network database of the service for geocoding, will automate the creation of map layers in a geographic information system based on text messages. The article describes the most popular approaches and their software implementations for solving the problem of extracting named entities in the example of texts of biographies and works of Siberian writers. Rule-based methodologies, maximum entropy models, and convolutional neural networks are analyzed. To assess the quality of the results of extracting geographical names and objects from the text, in addition to the standard F1-score, the authors propose an additional variant of the evaluation method that takes into account a larger number of criteria and is also based on an error matrix. The description of text block markup formats is given to improve the quality of recognition and expand the possible options for geographical names of named entities based on additional training of the neural network model.