Інтелектуальна технологія видобування та верифікації числових та текстових даних у багатозв’язних багатостадійних геоінформаційних системах

Abstract

The article is devoted to solving the problem of increasing the volume and level of reliability of information regarding the state of elements of multi-connected multi-stage geographic information systems (GIS) using information and intellectual technologies, which is especially relevant when monitoring the state of these objects. To solve the problem, numerical and textual data various sources (reports, cadasters, publications in the media and social networks, etc.) are used, which were bound and verified to the elements of the GIS taking into account spatial -temporal metadata. The criterion of topological observability is applied as a measure of the completeness of knowledge for the verification of numerical data in texts. For the first time, a method of verification of numerical data in texts using a knowledge base is proposed, the completeness of which is assessed using the criterion of topological observability, which makes it possible to assess the completeness and accuracy of the extracted information. The method of forming user queries to a large language model and RAG architecture has been improved by binding embeddings to spatial objects and temporal boundaries, which provides an opportunity to more accurately find information relevant to the user, taking into account the established spatial-temporal restrictions. The georeferencing method has been further developed, which automatically s relevant facts in texts and more accurately compares these facts with descriptions of water bodies. The integrated technology has been proposed that combines these methods — automated extraction of numerical and text facts various sources, their spatial-temporal binding to GIS elements and formalized verification in order to reduce information uncertainty and improve the quality of digital twins for decision support. An example of the application of this technology is given on the example of the task of analyzing the state of the water body of the Southern Bug River basin. The study showed that the reliability index of news about water pollution can be increased 0.3 to approximately 0.99, i.e. 3.3 times.