Aerospace Technic and Technology

The subject of the article is the methodology for building an integrated neuro-symbolic architecture of mental models for users with different levels of technical competence to generate personalized explanations of intelligent information systems decisions. The goal is to develop an architecture that provides automated detection of individual mental models from user behavioral data and the creation of interpretable symbolic representations of causal relationships with the adaptation of the level to detail to the user's skill level. The tasks addressed include: performing a comparative analysis of existing approaches to building mental models according to personalization criteria; developing an integrated neuro-symbolic architecture with functional distribution between neural network and symbolic components; conducting experimental verification of the proposed architecture; determining the scope of application for the developed architecture. The methods used include variational autoencoders with multi-channel attention mechanisms, neuro-symbolic translation with multi-level abstraction, and the generation of directed acyclic graphs. The following results were obtained: an integrated neuro-symbolic architecture was developed featuring a neural network component for the automated detection of individual cognitive structures through variational encoding and attention mechanisms with dynamic channel prioritization by user category, and a symbolic component for transforming latent descriptors into interpretable causal graphs with adaptive detailing and temporal validation to eliminate spurious dependencies. Conclusions. The results of the study confirmed the effectiveness of the integrated neuro-symbolic approach to building personalized mental models with the automated detection of latent cognitive structures from behavioral data without the need for expert knowledge. The scientific novelty of the results obtained lies in the development of an integrated neuro-symbolic architecture model that provides interaction between a neural network layer, designed for projecting behavioral trajectories into latent space and selecting significant features through multi-channel attention, and a symbolic layer for the neuro-symbolic translation of latent vectors into multi-level symbolic representation. This involves the generation of directed acyclic graphs and temporal validation, which improves the comprehensibility of intelligent systems' decisions through the construction of personalized, interpretable mental models. This, in turn, increases user trust in intelligent systems' decisions by tailoring explanations according to their level of technical competence.

neuro-symbolic architecture; mental models; personalized explanations; explainable artificial intelligence; variational autoencoders; attention mechanisms; causal graphs; temporal validation; user differentiation; interpretability

Kingma, D.P., & Welling, M. Auto-Encoding Variational Bayes. In Advances in Neural Information Processing Systems (NeurIPS 2013), vol. 27. Available at: https://arxiv.org/abs/1312.6114 (Accessed: 12 December 2025).

Badreddine, S., d’Avila Garcez, A., Serafini, L., & Spranger, M. Logic Tensor Networks. Artificial Intelligence, 2022, vol. 303, Article no. 103649. DOI:10.1016/j.artint.2021.103649.

Miller, T. Explanation in Artificial Intelligence: Insights from the Social Sciences. Artificial Intelligence, 2019, vol. 267, pp. 1–38. DOI: 10.1016/j.artint.2018.07.007.

Chromik, M., Eiband, M., Buchner, F., Krüger, A., & Butz, A. I Think I Get Your Point, AI! The Illusion of Explanatory Depth in Explainable AI’, in Proceedings of the 26th International Conference on Intelligent User Interfaces. New York, ACM, 2021, pp. 307–317. DOI:10.1145/3397481.3450644.

Nimmo, R., Constantinides, M., Zhou, K., Quercia, D., & Stumpf, S. User Characteristics in Explainable AI: The Rabbit Hole of Personalization? In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI 2024). 2024, New York, ACM. DOI: 10.1145/3613904.3642352.

Pearl, J. Causality: Models, Reasoning, and Inference. 2nd edn. Cambridge, Cambridge University Press, 2009. ISBN 978-0521895606.

Holzinger, A., Langs, G., Denk, H., Zatloukal, K., & Müller, H. Causability and Explainability of Artificial Intelligence in Medicine. WIREs Data Mining and Knowledge Discovery, 2019, vol. 9, iss. 4, article no. e1312. DOI:10.1002/widm.1312.

European Commission Proposal for a Regulation of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act). EUR-Lex, 2021. Available at: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52021PC0206 (Accessed: 12 December 2025).

Arrieta, A. B., Díaz-Rodríguez, N., Del Ser, J., & et al. Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI. Information Fusion, 2020, vol. 58, pp. 82–115. DOI:10.1016/j.inffus.2019.12.012.

Ribeiro, M. T., Singh, S., & Guestrin, C. “Why Should I Trust You?”: Explaining the Predictions of Any Classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York, ACM, 2016, pp. 1135–1144. DOI:10.1145/2939672.2939778.

Samek, W., Montavon, G., Lapuschkin, S., Anders, C. J., & Müller, K.-R. Explaining Deep Neural Networks and Beyond: A Review of Methods and Applications. Proceedings of the IEEE, 2021, vol. 109, iss. 3, pp. 247–278. DOI: 10.1109/JPROC.2021.3060483.

Prenosil, G. A., Weitzel, T. K., Afshar-Oromieh, A., & et al. Neuro-symbolic AI for auditable cognitive information extraction from medical reports. Communications Medicine, 2025, vol. 5, Article no. 491. DOI: 10.1038/s43856-025-01194-x.