Method of Text Summarization Using Graph Neural Networks

In today’s world large-scale language models (LLM) are gaining increasing popularity. These types of models are able to assist users in processing large volumes of text and performing tasks such as generating and classifying material, detecting plagiarism, and machine translation. They also can paraphrase texts and search for basic text information. Abstract systems based on neural networks enable quick extraction of key information from text documents, reducing their volume and making them more accessible for rapid familiarization and understanding. One such method for generating short content involves graph neural networks (GNNs), which take into account complex relationships between text concepts and sentences by representing textual information as a graph. The proposed solution is a sound mathematical method for text summarization, as well as software that implements a previously developed method for highlighting the main ideas of text materials. It also evaluates the methods performance. The experimental results indicate that the graph-based approach, using a graph model, has several advantages over traditional NLP approaches. This makes it more convenient when working with large volumes of text. The methods described in this paper can be used to generate annotations for articles, educational materials, and fiction. Therefore, the development of a text summary method based on graph neural networks has significant practical importance and can enhance the quality of automated text compression.