Semantic Annotation of Russian-Language Tables Based on a Pre-Trained Language Model

Tables are widely used in organizing, presenting, and storing data. However, as a rule, they have different arbitrary layouts and are not accompanied by explicit semantics necessary for machine interpretation of their content. Recent research has shown great progress in solving various problems in the field of automatic table understanding. Despite this, there is still a need to develop more effective solutions that are applicable in practice and take into account data with different linguistic backgrounds. In this paper, we propose a new RuTaBERT framework based on a pre-trained language model and provide semantic column annotation of Russian-language tables. We used a preprocessed large-scale corpus of Russian-language web tables extracted from Wikipedia pages as training data to fine-tune the language model. The experimental results obtained show the prospect of using the proposed framework, including when solving applied problems.