R.E. Semenov1
1 MIREA – Russian Technological University (Moscow, Russia)

1 9629790@gmail.com

The development of a neural conceptual-graph network for text understanding is driven by the need to create more efficient methods for text analysis that provide a deep understanding of their content and meaning, even with a limited volume of training data. The Russian language is a complex system that presents unique challenges: complex morphology with numerous declension and conjugation forms, intricate syntax, a rich vocabulary with many synonyms and homonyms, a large number of idiomatic expressions, as well as a variety of stylistic forms and regional features. These aspects require the development of neural networks capable of accounting for linguistic nuances.

Existing approaches to text understanding include analytical-synthetic models, which identify key elements of a text and transform them into structures such as summaries or tables. These models often rely on pre-trained databases to analyze new information and assess its validity. Hermeneutic models interpret texts through the lens of the author's ideas, revealing the motivation and context behind their creation.

The conceptual-graph approach focuses on identifying key elements of a text by representing it as a graph, where the nodes correspond to concepts and the edges represent relationships between them. This structure allows for modeling text on multiple levels. On the morphological level, parts of speech that are significant for text comprehension are analyzed, such as nouns, verbs, adjectives, participles, and pronouns. On the syntactic level, sentence structures and semantic relationships are examined, including subject, object, actions, and their characteristics. The semantic level organizes elements of the text in terms of the main idea, description, and logical connections, while the pragmatic level considers the goals and audience of the text.

The efficiency of the conceptual-graph model is significantly enhanced by the use of sparse data, enabling the processing of large volumes of text with minimal loss of information. This approach eliminates secondary constructions, focusing attention on the significant parts of the text. The results of the conducted experiments showed that the conceptual-graph model achieves high accuracy when working with small datasets, outperforming existing algorithms such as GPT, BERT, and ELMo. However, as the volume of data increases, the model's accuracy decreases due to the complexity of integrating various algorithms.

The conceptual-graph neural network is a promising approach in the field of natural language processing, providing efficient analysis of texts across different genres and styles. This method allows for the visualization of key semantic relationships and the creation of more accurate neural networks for text information analysis.

Semenov R.E. Neural conceptual-graph network on a limited dataset. Neurocomputers. 2025. V. 27. № 1. P. 28–38. DOI: https://doi.org/ 10.18127/j19998554-202501-02 (in Russian)

1. Sadovskaya L.L., Gus'kov A.E., Kosyakov D.V., Mukhamediev R.I. Obrabotka tekstov na estestvennom yazyke: obzor publikatsij. Iskusstvennyj intellekt i prinyatie reshenij. 2021. № 3. S. 66–86. (in Russian)
2. Sorokin A.B., Lobanov D.A. Kontseptual'noe proektirovanie intellektual'nykh system. Informatsionnye tekhnologii. 2018. T. 24. № 1. S. 3–10. (in Russian)
3. Liu B., Wu L. Graph neural networks in natural language processing. Ch. 21 in Graph Neural Networks: Foundations, Frontiers, and Applications. Ed. by L. Wu et al. Springer Nature Singapore Ltd. 2022. P. 463–481.
4. Tsitul'skij A.M., Ivannikov A.V., Rogov I.S. NLP-obrabotka estestvennykh yazykov. Nauchno-obrazovatel'nyj zhurnal dlya studentov i prepodavatelej «StudNet». 2020. № 6. S. 467–475. (in Russian)
5. Anferov M.A. Geneticheskij algoritm klasterizatsii. Russian Technological Journal. 2019. T. 6. № 7. S. 134–150. (in Russian)
6. Zhang X., Zhao H., Chen D.-Y. Semantic mapping methods between expert view and ontology view. Journal of Software. 2020. V. 31. № 9. P. 2855–2882.
7. Sorokin A.B., Smol'yaninova V.A. Kontseptual'noe proektirovanie ekspertnykh sistem podderzhki prinyatiya reshenij. Informatsionnye tekhnologii. 2017. T. 23. № 9. S. 634–641. (in Russian)
8. Tsitul'skij A.M., Rogov I.S., Ivannikov A.V. Intellektual'nyj analiz teksta. Nauchno-obrazovatel'nyj zhurnal dlya studentov i prepodavatelej «StudNet». 2020. № 6. S. 476–483. (in Russian)
9. Sorokin A.B., Zheleznyak L.M., Suprunenko D.V., Kholmogorov V.V. Proektirovanie modulej sistemnoj dinamiki v sistemakh podderzhki prinyatiya reshenij. Russian Technological Journal. 2022. T. 10. № 4. S. 18–26. (in Russian)
10. Ol'gina I.G., Pronin I.V., Abdrakhmanov A.N. Postroenie grafovykh modelej seti tsitirovaniya nauchnykh publikatsij. Sistemy upravleniya, informatsionnye tekhnologii i matematicheskoe modelirovanie. 2020. № 1. S. 118–125. (in Russian)
11. Anferov M.A. Algoritm poiska podkriticheskikh putej na setevykh grafikakh. Russian Technological Journal. 2023. T. 11. № 1. S. 60–69. (in Russian)
12. Krasnikov K.E. Matematicheskoe modelirovanie nekotorykh sotsial'nykh protsessov s pomoshch'yu teoretiko-igrovykh podkhodov i prinyatie na ikh osnove upravlencheskikh reshenij. Russian Technological Journal. 2021. T. 9. № 5. S. 67–83. (in Russian)
13. Tomashevskaya V.S., Yakovlev D.A. Sposoby obrabotki nestrukturirovannykh dannykh. Russian Technological Journal. 2021. T. 9. № 1. S. 7–17. (in Russian)
14. Tatur M.M., Lukashevich M.M., Pertsev D.Yu., Iskra N.A. Intellektual'nyj analiz dannykh i oblachnye vychisleniya. Doklady BGUIR. 2019. № 6(124). S. 62–71. (in Russian)
15. Minaev V.A., Simonov A.V. Sravnenie modelej-transformerov bert pri vyyavlenii destruktivnogo kontenta v sotsial'nykh media. Informatsiya i bezopasnost'. 2022. T. 25. № 3. S. 341–348. (in Russian)
16. Kochkarov R.A. Issledovanie nekotorykh trudnorazreshimykh zadach na klasse predfraktal'nykh grafov s izmenyaemym traektornym porozhdeniem. Vestnik Voronezhskogo gosudarstvennogo universiteta. Seriya: Sistemnyj analiz i informatsionnye tekhnologii. 2021. №. 4. S. 66–82. (in Russian)
17. Ivanov M.V. Sovremennye nejrosetevye podkhody k resheniyu zadach na grafakh. Sb. materialov nauch.-praktich. konf. «Primenenie iskusstvennogo intellekta v informatsionno-telekommunikatsionnykh sistemakh». Sankt-Peterburg. 2021. S. 90–98. (in Russian)
18. Gorrec L., Knight Ph.A., Caen A. Learning network embeddings using small graphlets. Social Network Analysis and Mining. 2022. P. 12–20.
19. Belov S.D., Zrelova D.P., Zrelov P.V., Koren'kov V.V. Obzor metodov avtomaticheskoj obrabotki tekstov na estestvennom yazyke. Sistemnyj analiz v nauke i obrazovanii. 2020. № 3. S. 8–22. (in Russian)
20. Wang H., Li J., Wu H., Hovy E., Sun Y. Pre-trained language models and their applications. Engineering. 2023. V. 25. P. 51–65.