A hybrid method for extracting keywords from Russian-language scientific and technical texts based on syntactic analysis and multifactor ranking

Abstract

This is article is devoted to the development of a hybrid method for automatically extracting keywords from Russian-language scientific and technical texts. The proposed method combines morphosyntactic analysis, candidate generation based on syntactic dependency tree traversal, and multivariate ranking using a combination of statistical (TF-IDF, C-value), graph (TextRank), and positional features. An experimental evaluation on a corpus of 16,818 Russian-language scientific articles from CyberLeninka showed that the proposed method outperforms baseline algorithms (TF-IDF, RAKE, YAKE, TextRank) in terms of F1@10 (2.3% improvement – 0.268 versus 0.262 for the closest competitor TextRank), MAP (6.9% improvement – 0.217 versus 0.203), and semantic text coverage (32.5% improvement – 0.559 versus 0.422). A particularly significant advantage is achieved when extracting multi-word terminological phrases due to the consideration of syntactic relationships between components.