Algorithms and methods for automated construction of knowledge graphs based on text sources

Abstract

In this article, we present our path towards building knowledge graphs automatically from Russian texts. We explore various methodologies and libraries to extract triples, which are the fundamental building blocks of knowledge graphs. Our approach involves the use of libraries for analyzing morphological characteristics of words, such as PyMorphy and Yandex Mystem, to construct triples. We also utilize the NLP library spaCy to analyze text and build triples based on semantic relationships recognized by the library. However, we found that in some cases, we could not extract relationships from the text, leading us to use word2vec to define relationships. Unfortunately, the results obtained from word2vec were unsatisfactory and could not be used as relationships. We also encountered the problem of building triples from text due to the use of pronouns. To address this issue, we explored the use of coreference resolution libraries, but unfortunately, there are no working libraries available for the Russian language at this time. Our results highlight both positive and negative outcomes of applying these methodologies and libraries, providing insights into the challenges and opportunities of building knowledge graphs automatically from Russian texts.