A System for Automatic Construction of Knowledge Graphs of Mathematical Documents

This article outlines the process of creating an automated system for knowledge graph construction from collections of mathematical documents in LATEX format. The Math-CollectionOntology, which defines the types of objects and relationships in knowledge graphs, was developed. The introduced toolkit includes methods for extracting mathematical terms, browsing and identifying document topics, extracting entities from LATEX code, and calculating statistical parameters of the graph. The parsed entities are mathematical terms, topics generated through the Latent Dirichlet Allocation, UDC codes, used formulas, author affiliations, cited literature, and others. The knowledge graph captures each extracted object using specific types of relationships defined in the MathCollectionOntology. Here, a knowledge graph was coined for a collection of articles published in Izvestiya VUZov. Matematika journal (1114 Russian-language documents in LATEX format). The thematic terms of the document topics were described. The quantitative parameters of the constructed knowledge graph were obtained.