Resumen
The article discusses existing approaches to automatic text abstracting. The creation of an abstract obviously requires an understanding of the text at the level of pragmatics. However, at the moment, reliable semantic analysis is still not available for computers, especially not the analysis of pragmatics. Widespread methods based on neural networks can take into account semantics thanks to special vector representations of words. But the rest of the automatic referencing methods rely on morphology and syntax. However, part of the semantics is still available to them - there are thesauruses and networks, as well as algorithms that allow you to establish a semantic connection between individual words, such as their semantic similarity, in particular, synonymy. A method is proposed that takes into account the semantic similarity of words. A modification of the graph method has been developed that takes into account synonymy and allows for a better abstract. The basic and modified versions of the graph method are implemented programmatically for automatic referencing tasks, taking into account synonymy. The comparison of their quality in Russian and English texts using automatic evaluation metrics of abstracts is carried out. The evaluation results show an improvement in the modified graph method compared to the usual one, especially in Russian-language texts.