Abstract: The paper is devoted to the development of a hybrid system for creating digital twins of regulations, integrating neural network methods with formal logic models. Modern approaches based on embeddings and large language models have the disadvantage of "black box", which limits their application in critical areas. In this paper, we propose a five-stage algorithm including linguistic preprocessing, construction of atomic diagrams through the LogicText system, generation of vector embeddings, ontological comparison of vector and logical representations of information, and query correction. The system is implemented using the Spring Boot platform with the Camunda BPMN engine for automation of control of regulated processes of university departments. The hybrid approach addresses the opacity problem of language models, providing interpretability and formal verification while preserving the efficiency of neural network methods. The results demonstrate the feasibility of creating a digital deputy for the department secretary.

Cite: Palchunov D. , Yakobson A. , Nemtsev I. , Shelkovnikova S.
Integration of logical and neural network methods for creation of digital twins of regulations
In compilation IEEE XVII International Conference on Actual Problems of Electronic Instrument Engineering (IEEE APEIE 2025). 2025. DOI: 10.1109/apeie66761.2025.11289398 OpenAlex

Identifiers:

OpenAlex:

W4417472786

Citing: Пока нет цитирований

Altmetrics: