| Abstrakt: |
An overview of approaches to automatic extraction of taxonomic relationships (is_a relationships) from texts, including classical methods based on lexico-semantic patterns and vector representations of words and their modern development, is provided. It also describes new approaches that are based on language models like BERT and use new options for working with lexico-semantic patterns, in which the model must fill the desired position in the pattern. Various ways to test the quality of extraction of taxonomic relationships and the results of these tests are presented. According to the results currently achieved, it can be concluded that, despite the progress in the development of methods for extracting taxonomic relationships, the quality of extraction is insufficient for fully automatic construction or enrichment of taxonomies. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink