Methods and techniques to automatic entity linking in Russian

Nowadays, there is a growing interest in solving NLP tasks using external knowledge storage, for example, information retrieval, question-answering systems, dialogue etc. Thus it important to establish relations between entities the processed text and base. This article devoted entity linking, where Wikidata used as an We consider scientific terms Russian entities. Traditional linking system has three stages: recognition, candidates (from base) generation, candidate ranking. Our takes raw with defined input. To generate we use string match input from Wikidata. The ranking stage most complicated one because requires semantic information. Several experiments different models were conducted, including approach based on cosine similarity, classical machine learning algorithms, neural networks. Also, extended RUSERRC dataset, adding manually annotated data model training. results showed that similarity leads better compared others doesn’t require data. dataset are open-sourced available other researchers.