UDC 004.89:005.53
COMPARATIVE ANALYSIS OF ARCHITECTURES OF DECISION SUPPORT SYSTEMS BASED ON RULES, LARGE LANGUAGE MODELS, AND THEIR HYBRID COMBINATIONS
A. O. Kostyrenkov, PhD Student, Assistant Professor, Department of Instrumentation and Applied Software, IIT RTU MIREA, Moscow, Russia;
orcid.org/0009-0007-0294-694X, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
E. N. Miftakhov, Doctor in Physics and Mathematics, professor, Department of Instrumentation and Applied Software, IIT RTU MIREA, Moscow, Russia;
orcid.org/0000-0002-0471-5949, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
This paper presents a comparative analysis of three decision support system (DSS) architectures for de vice control: rule-based, LLM-based, and hybrid (LLM + rules). Particular attention is paid to exploring the balance between the predictability of actions and the security of rules, as well as the flexibility of natural language understanding in large language models (LLM) susceptible to hallucinations and prompt attacks. For comparison, simplified prototypes of all three architectures were implemented, and an experiment was conducted on the same set of Russian-language commands with multiple repetitions. The accuracy of opera tion, correct execution of valid commands, the justification for rejecting invalid commands, response time, and resilience to wording variability were assessed, allowing us to evaluate the effectiveness of each archi tecture
Key words: Large language models (LLM), hybrid architecture, neuro-symbolic approach, decision support systems (DSS), IoT system.