Отримано 02.01.2025, Доопрацьовано 11.03.2025, Прийнято 24.04.2025

Метод побудови когнітивної карти процесів у динамічній системі із використанням кооперації великих мовних моделей

Борис Варер , Віталій Мокін

В умовах постійного зростання вимог до швидкого прийняття рішень і глибокого аналізу складних динамічних систем, коли доступні дані обмежені, а залучення досвідчених експертів часто є неможливим або занадто витратним, розроблення нових методів побудови моделей набуває особливої актуальності. Використання великих мовних моделей (LLM) як експертних систем дозволяє суттєво знизити ресурсні витрати та прискорити процес моделювання складних технічних, екологічних та соціально-економічних систем. Метою даної роботи було дослідження та практична демонстрація потенціалу та можливостей LLM, як експертних систем, у процесі побудови когнітивних карт. У даній роботі запропоновано та обґрунтовано архітектуру кооперації ансамблів LLM для формалізованого генерування вершин-змінних та вагових коефіцієнтів когнітивних карт, що дозволяє автоматизувати процес моделювання без залучення експертів-людей. Здійснено декомпозицію типового промпта (вказівки) до LLM на структурні складові: опис контексту (D), рольову настанову моделі (R), інструкцію (I), умови (C) та формат відповіді (F) та запропоновано підхід їх визначення експертним шляхом. Розроблено систему таких промптів, яка забезпечує структуроване оброблення даних та ідентифікацію взаємозв’язків між елементами системи. Практичну ефективність підходу продемонстровано на прикладі прогнозування стану води у Сабарівському водосховищі біля м. Вінниця, де для більшості фізико-хімічних показників моделювання продемонструвало малу похибку (2.09–4.60 %) навіть за мінімального обсягу вхідних даних. Запропонований метод є перспективним для задач моделювання та прогнозування у складних системах з обмеженим обсягом даних, зокрема в екологічних, соціально-економічних та інженерних сферах, де швидкість отримання надійних результатів має критичне значення для прийняття обґрунтованих рішень

ВММ; генеративний штучний інтелект; інтелектуальна технологія; системний аналіз; моделювання; прогнозування; динамічна система
69-78
Varer, B., & Mokin, V. (2025). Method for constructing a cognitive map of processes in a dynamic system using the cooperation of large language models. Information Technologies and Computer Engineering, 22(1), 69-78. https://doi.org/10.63341/vitce/1.2025.69

Використані джерела

[1] Akiba, T., Shing, M., Tang, Y. & Ha, D. (2025). Evolutionary optimization of model merging recipes. Nature Machine Intelligence, 7(2), 195-204. doi: 10.1038/s42256-024-00975-8.

[2] Bender, E.M., & Koller, A. (2020). Climbing towards NLU: On meaning, form, and understanding in the age of data. In Proceedings of the 58th annual meeting of the association for computational linguistics (pp. 5185-5198). Stroudsburg: Association for Computational Linguistics. doi: 10.18653/v1/2020.acl-main.463.

[3] Bender, E.M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency (FAccT‘21) (pp. 610-623). Stroudsburg: Association for Computing Machinery. doi: 10.1145/3442188.3445922.

[4] Cao, H., Ma, R., Zhai, Y., & Shen, J. (2024). LLM-Collab: A framework for enhancing task planning via chain-of-thought and multi-agent collaboration. Applied Computing and Intelligence, 4(2), 328-348. doi: 10.3934/aci.2024019.

[5] Chen, S.-W., & Hsu, H.-J. (2023). MisCaltral: Reducing numeric hallucinations of mistral with precision numeric calculation. Research Square. doi: 10.21203/rs.3.rs-3789011/v1.

[6] Cherniuk, O. (2023). Modeling the impact of AI-based chatbots on the quality of higher education using system analysis methods. Information Technologies and Society, 3(9), 80-90. doi: 10.32689/maup.it.2023.3.11.

[7] Das, S., & Srihari, R. (2024). Compos mentis at SemEval2024 Task6: A multi-faceted role-based large language model ensemble to detect hallucination. In Proceedings of the 18th international workshop on semantic evaluation (SemEval-2024) (pp. 1449-1454). Mexico City: Association for Computational Linguistics. doi: 10.18653/v1/2024. semeval-1.208.

[8] Feleki, A., Apostolopoulos, I.D., Moustakidis, S., Papageorgiou, E.I., Papathanasiou, N., Apostolopoulos, D., & Papandrianos, N. (2023). Explainable deep fuzzy cognitive map diagnosis of coronary artery disease: Integrating myocardial perfusion imaging, clinical data, and natural language insights. Applied Sciences, 13(21), article number 11953. doi: 10.3390/app132111953.

[9] Godoy, W.F., Fabri, J.A., Palácios, R.H.C., Mendonça, M., Gonçalves, J.F.S., & Moraes, L.O.M. (2024). Using fuzzy cognitive maps and chatbots to evaluate student satisfaction in a university: A comparison between strong and weak AI. In Proceedings of the eighteenth international conference on mobile ubiquitous computing, systems, services and technologies (pp. 16-20). Wilmington: IARIA Press.

[10] Huang, L., et al. (2025). A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems, 43(2), article number 42. doi: 10.1145/3703155.

[11] Kong, A., Zhao, S., Chen, H., Li, Q., Qin, Y., Sun, R., Zhou, X., Wang, E., & Dong, X. (2024). Better zero-shot reasoning with role-play prompting. In Proceedings of the 2024 conference of the North American chapter of the association for computational linguistics: Human language technologies (Vol. 1, pp. 4099-4113). Mexico City: Association for Computational Linguistics. doi: 10.18653/v1/2024.naacl-long.228.

[12] Liu, T.J., Boulle, N., Sarfati, R., & Earls, C. (2024). LLMs learn governing principles of dynamical systems, revealing an in-context neural scaling law. In Proceedings of the 2024 conference on empirical methods in natural language processing (pp. 15097-15117). doi: 10.18653/v1/2024.emnlp-main.842.

[13] Lu, J., Pang, Z., Xiao, M., Zhu, Y., Xia, R., & Zhang, J. (2024). Merge, ensemble, and cooperate! A survey on collaborative strategies in the era of large language models. ArXiv. doi: 10.48550/arXiv.2407.06089.

[14] Mokin, V.B., Burdeina, O.V., & Varchuk, I.V. (2020). On the optimization of topologically observable cognitive maps while preserving their robustness. Visnyk of Vinnytsia Polytechnic Institute, (6), 84-92. doi: 10.31649/1997-9266-2020153-6-84-92.

[15] Mokin, V.B., Dratovanyi, M.V., Kozachko, O.M., & Zhukov, S.O. (2021). Method for synthesizing a robust multi-connected cognitive map of a complex system. Visnyk of Vinnytsia Polytechnic Institute, (6), 114-122. doi: 10.31649/1997-92662021-159-6-114-122.

[16] Roberts, F. (1976). Discrete mathematical models with applications to social, biological, and environmental problems. Englewood Cliffs: Prentice-Hall.

[17] Romanenko, V., & Miliavskyi, Y. (2022). Coordinating control of a cognitive map impulse process in stochastic environment. Problems of Control and Informatics, 67(4), 49-58. doi: 10.34229/2786-6505-2022-4-4.

[18] Saliieva, O., & Yaremchuk, Yu. (2020). Study of the reliability of the impact of threats on the level of security of the information protection system and the object of critical infrastructure based on the results of cognitive modeling . Bulletin of Cherkasy State Technological University, 25(3), 85-93. doi: 10.24025/2306-4412.3.2020.216251.

[19] Schoenegger, P., Tuminauskaite, I., Park, P.S., Bastos, R.V.S., & Tetlock P.E. (2024). Wisdom of the silicon crowd: LLM ensemble prediction capabilities rival human crowd accuracy. Science Advances, 10(45), article number eadp1528. doi: 10.1126/sciadv.adp1528.

[20] Schuerkamp, R., Ahlstrom, H., & Giabbanelli, P.J. (2025). Automatically resolving conflicts between expert systems: An experimental approach using large language models and fuzzy cognitive maps from participatory modeling studies. Knowledge-Based Systems, 313, article number 113151. doi: 10.1016/j.knosys.2025.113151.

[21] Shevchenko, S., Zhdanova, Y.A., Kryvytska, O., Shevchenko, H., & Spasiteleva, S. (2024). Fuzzy cognitive mapping as a scenario approach for information security risk analysis (short paper). In Proceedings of the 2024 cybersecurity providing in information and telecommunication systems II (pp. 356-362). Kyiv: Borys Grinchenko Kyiv Metropolitan University.

[22] Vinnytsia City Council. (2024). Ecology and natural resources. Retrieved from https://www.vmr.gov.ua/ecology.

[23] Wang, Z.M., et al. (2024). RoleLLM: Benchmarking, eliciting, and enhancing role-playing abilities of large language models. In Findings of the Association for computational linguistics (ACL 2024) (pp. 14743-14777). Bangkok: Association for Computational Linguistics. doi: 10.18653/v1/2024.findings-acl.878.