Extraction of relevant lexis has gained significance as the
amount of information is continuously growing with news, posts on social networks, reviews, academic papers, etc. piling up. Automated algorithms are needed to analyze texts to facilitate understanding of their content. The paper scrutinizes methods for keyword extraction in abstracts
of Russian scientific texts on computational linguistics. Unsupervised algorithms based on statistics, graphs and machine learning principles are
considered. The results are evaluated against the keywords assigned by
authors themselves, followed by expert opinion. Log-likelihood produced
the best results in comparison with author keywords, while KeyBERT implementation with vectorizers outperformed other algorithms according
to expert assessment.