INTERPRETABILITY OF NEUROSEMANTIC MODELS IN THEIR APPLICATION IN APPLIED FIELDS

K. A. Nikitenko, A. V. Zviagintseva

About the authors

K. A. Nikitenko, A. V. Zviagintseva

Никитенко Кирилл Андреевич – аспирант кафедр ы компьютерных технологий ФГБОУ ВО «ДонГУ», 283001, Донецк, ул. Университетская, 24. Область научных интересов: компьютерное зрение, машинное обучение, нейронные сети. Число научных публикаций – 5.

Звягинцева Анна Викторовна – д.т.н. доцент, профессор кафедры компью- терных технологий ФГБОУ ВО «ДонГУ», 283001, Донецк, ул. Университетская, 24. Область научных интересов: системный анализ, событийная и комплексная оценка; безопасность и управление социально -экономическими и техногенными системами; информационно-аналитические системы; обработка и анализ данных. Число научных публикаций – более 150.

Download PDF

UDC: 004.8+004.912
DOI: 10.24412/2413-7383-2025-2-37-79-90
Language: Russian

Abstract: The article addresses the problem of interpretability in neurosemantic models used for natural language processing tasks. It discusses the main challenges related to the interpretation of high - dimensional vector representations, context -dependent features, and deep transformer archi - tectures. The paper outlines current approaches to interpretability, including both intrinsic and post - hoc methods. Special attention is given to the trade-off between accuracy and interpretability, as well as to the prospects for further development in this area.

Keywords: interpretability, neurosemantic models, applied areas, transformer, explainability, practical application.

References

1. Model neuroseteum: Chto takoe I Chem otlichayutsya drug OT druga // unisender: site – 2025 (accessed May 2, 2025).
2. Amodei D. Concrete Problems in AI Safety. // archive. 2016. URL: https://arxiv.org/pdf/1606.06565 (accessed May 2, 2025).
3. Jarowa M. Model BERT dlya mashinnogo obucheniya // habr. 2024. URL: https://habr.com/ ru/companies/skillfactory/articles/862130 (accessed May 28, 2025).
4. ChatGPT 2025 // habr-2025, URL: https://habr.com/ru/companies/bothub/articles/887988/ (accessed May 28, 2025).
5. BioBERT-model obrabotki biomedisinskix tekstov // neurohive. 2019, URL: https://neurohive.io/ ru/novosti/biobert/ (accessed May 28, 2025).
6. ClinicBERT // HuggingFace -2023, URL: https://huggingface.co/tdobrxl/ClinicBERT (accessed May 2, 2025).
7. RuT5, RuRoBERTa, RuBERT: Kak Mi obuchili seriyu modeley dlya russkogo yazika // habr – 2021, URL: https://habr.com/ru/companies/sberdevices/articles/567776/ (accessed May 28, 2025).
8. Distillbert: the compact NLP Powerhouse // opengenus-2023, URL: https://iq.opengenus.org/distilbert/ (accessed May 28, 2025).
9. XLM-RoBERTa // huggingface, URL: < BR > https://huggingface.co/docs/transformers/model_doc/xlm- roberta (accessed May 28, 2025).
10. Beltagy L. Lo K. Cohan A. SciBERT: a pretrained languge model for scientific text // archive – 2019, URL: < br > https://huggingface.co/docs/transformers/model_doc/xlm-roberta (accessed May 28, 2025).
11. T. Pires, E. Schlinger, D. Garrett, How multilingual is mBERT? // archive-2020, URL: https://arxiv.org/ abs/2005.09093 (accessed May 28, 2025).
12. Reformer-effektivniy transformer // habr. 2020, URL: https://habr.com/ru/articles/522622/ (accessed May 29, 2025).
13. L. Beltagy, M. Peters, A. Cohan, LongFormer: the long-document Transformer // archive – 2020, URL: < BR > https://arxiv.org/abs/2004.05150 (accessed May 29, 2025).
14. F. Landola, A. Shaw, R. Krishna, K. Keutzer, SqueezeBERT: what can computer vision teach NLP about efficient neural networks? // archive-2020, URL: https://arxiv.org/abs/2006.11316 (accessed May 29, 2025).
15. W. Wang, et al. MiniLM: Self Attention distillation for task-agnostic compression of pre-trained Transformers // archives. 2020, URL: < BR > https://arxiv.org/abs/2002.10957 (accessed May 29, 2025).
16. Google T5 1.1 model // metatext-2024, URL: https://metatext.io/models/google-t5-v1_1-xxl (accessed May 29, 2025).
17. Novaya yazikovaya model Google PalM // habr-2022, URL: https://habr.com/ru/news/659603/ (accessed May 28, 2025).
18. Yandex vilojil YaLM-krupneyshaya GPT-podobnaya neuroset // habr. 2022, URL: https://habr.com/ru/companies/yandex/articles/672396/ (accessed May 28, 2025).
19. Lepikhin D. et al. GShard: Scaling giant models with conditional computing and automatic Sharding // archive. 2020, URL: < BR > https://arxiv.org/abs/2006.16668 (accessed May 29, 2025).
20. GODEL: mojet Lee chatbot rabotat v korotkix besedax? Mycrosoft Govorit da! // analyticsindiamag-2022, URL: https://analyticsindiamag.com/global-tech/can-a-chatbot-indulge-in-small-talks-microsoft-says-yes/ (accessed May 29, 2025).
21. DialoGPT – a state of the art large-scale pretrained response generation model // huggingface, URL: < br > https://huggingface.co/microsoft/DialoGPT-medium (accessed May 29, 2025).
22. Zhao H. et al. Explainability for Large Language Models: a Survey // archive preprint. 2024. archive: 2309.01029. URL: https://arxiv.org/pdf/2309.01029 (accessed May 29, 2025).
23. Räuker T. Ho A. Casper S. Hadfield-Menell D. Toward Transparent AI:a Survey on Interpreting the Inner Structures of Deep Neural Networks // archive: site. – 2022, URL: https://arxiv.org/pdf/2207.13243 (accessed May 28, 2025).
24. Xia B. Wang X. Yamasaki T. Semantic exploration for Deep Neural Networks Using Feature Interactions // ACM Transactions on Multimedia Computing, Communications, and Applications. V.17, no 2, 2021: 1 –
25. DOI: 10.1145/3474557.
26. Ermolenko T.V. Classification oshibok V tekste na osnove glubokogo obucheniya // problemi iskusstvennogo intellekta. V.3, no 14, 2019: 47 –57.
27. Smirnov I.V. Intellectualny analiz tekstov na osnove metodov raznourovnevoy obrabotki estestvennogo yazika. Moscow, FITS IO RAN, 2023, 354 p. Интерпретируемость нейросемантических моделей при их применении… Н
28. Sun X. et al. Interpreting Deep Learning Models in Natural Language Processing: a Review // archive, 2021. archive: 2110.10470. URL: https://arxiv.org/pdf/2110.10470 (accessed May 29, 2025).
29. Ji Y. et al. A Comprehensive Survey on Self-Interpretable Neural Networks // archive, 2025. archive: 2501.15638. URL: https://arxiv.org/pdf/2501.15638 (accessed May 2, 2025).
30. Antsiferov S.S. Methodology razvitia intellectualnix system/ s.S. Antsiferov, A.S. Sigov, K.N. Fazilova. Problemi iskusstvennogo intellekta, no 2(25), 2022: 42 –47.
31. Beresca L. Gavves S.. Mechanistic Interpretability for AI Safety // OpenReview: site, URL: https://arxiv.org/pdf/2501.15638 (accessed May 29, 2025).
32. Starchenko S.N. Rudakov A.V. Postroenie interpretiruemix modeley dlya analiza polzovatelskix otzivov // Informasionnie technologii I vichislitelnie system, no 3, 2020: 73 –83.
33. Luber M. Thielmann A. Safken B. Structural Neural Aditive Models: enhanced Interpretable Machine Learning / / archive: site, URL: https://arxiv.org/pdf/2302.09275 (accessed May 2, 2025).
34. Madsen A. Reddy S. Chandar S. Post-Hoc Interpretability for Neural NLP: a Survey // ACM Computing Surveys, no 5(155), 2022: 1 –42.
35. Kiselyov S.A. Architekturi sovremennix yazikovix modeley: OT BERT do GPT-3 // Sistemi I sredstva informatiki, V.33, no 2, 2023: 95 –114.
36. Melnikova N.N. Shklyar V.L. Method interpretasii resheniy neurosetevix modelei V zadachax analiza texta / / problemi programmirovaniya, no 5, 2021: 102 –110.
37. Zhuraev I.I. Popov S.V. Interpretiruemost algorithmov mashinnogo obucheniya V zadachax mediisini / / iskusstvenniy intellect I prinyatie resheniy, no 1, 2022: 54 –64.
38. Chuprov I.Yu. Fyodorov V.A. Interpretiruemost neurosetevix modeley: problemi I podhodi // Vestnik Moskovskogo universiteta. Series 1: Mathematics I mechanics, no 6, 2023: 41 –56.

Issue: 2(37)'2025
Section: SYSTEM ANALYSIS, CONTROL AND INFORMATION PROCESSING, STATISTICS
How to cite: K. A. Nikitenko, A. V. Zviagintseva. INTERPRETABILITY OF NEUROSEMANTIC MODELS IN THEIR APPLICATION IN APPLIED FIELDS // Problems of Artificial Intelligence. - 2025. - № 2 (37). - P. 79-90. - https://paijournal.guiaidn.ru/en/2025/2(37)-8.html

Back to issue 2(37)'2025