E.V. Romanova1, V.A. Malekova2, I.V. Menshikov3

1–3 University Financial University under the Government of the Russian Federation (Moscow, Russia)
1 ekvromanova@fa.ru, 2 vamalekova@fa.ru, 3 216900@edu.fa.ru

Modern development of artificial intelligence technologies, especially in the field of natural language processing, provides significant prospects for the automation of processes of analysis and processing of large volumes of digital information. Given the exponential growth of data volumes and rising expectations for rapid access to relevant and reliable information, the creation of intelligent systems capable of quickly and accurately responding to user requests using specialized knowledge is becoming increasingly in demand.

The aim of this research is to develop an intelligent system for automatic generation of responses to typical user queries based on RAG and transformer technologies, supplemented by an integrated knowledge graph to optimize the process.

As a result, the principles of operation of transformers and the RAG method were studied; a method for assessing the quality of generated responses was developed; methods for improving generation by integrating knowledge graphs were investigated; a prototype of the system was implemented and designed as a practical product – a Telegram bot.

A chatbot architecture for Telegram is also proposed, combining the classic Retrieval-Augmented Generation (RAG) method with the integration of a knowledge graph using the LightRAG scheme – an effective tool for automating routine document processing operations and communication with clients. A comparison was conducted which showed that the proposed approach outperforms the classical RAG in terms of key quality metrics: answer reliability, completeness and accuracy of context, ranking relevance, and METEOR and BERTScore scores. The developed chatbot demonstrates high efficiency and flexibility. The study confirms the feasibility of integrating graph structures with RAG for tasks requiring a balanced synthesis of detailed and contextual knowledge.

Romanova E.V., Malekova V.A., Menshikov I.V. Development of a chatbot for question answering using knowledge graphs and retrieval-augmented generation. Dynamics of complex systems. 2025. V. 19. № 5 P. 77−86. DOI: 10.18127/j19997493-202505-09 (in Russian).

1. Lewis P., et al. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. arXiv preprint arXiv:2005.11401. 2020. URL: https://arxiv.org/abs/2005.11401. (data obrashheniya: 10.05.2025). CSS. URL: https://yoksel.github.io/size-units (data obrashheniya: 04.12.2022).
2. Edge D., et al. From Local to Global: A Graph RAG Approach to Query-Focused Summarization. arXiv preprint arXiv:2404.16130. 2024. URL: https://arxiv.org/abs/2404.16130. (data obrashheniya: 10.06.2025).
3. Guo Z., et al. LightRAG: Simple and Fast Retrieval-Augmented Generation. arXiv preprint arXiv:2410.05779. 2025. URL: https://arxiv.org/ abs/2410.05779 (data obrashheniya: 27.05.2025).
4. Global Artificial Intelligence Market Size Outlook. Grand View Research. E`lektronny`j resurs. URL: https://www.grandviewresearch.com/horizon/ outlook/artificial-intelligence-market-size/global (data obrashheniya: 08.06.2025).
5. Key Chatbot Statistics. Botpress. E`lektronny`j resurs. URL: https://botpress.com/blog/key-chatbot-statistics#:~:text=1 (data obrashheniya: 08.06.2025).
6. RAG Explained. SuperAnnotate. E`lektronny`j resurs. URL: https://www.superannotate.com/blog/rag-explained (data obrashheniya: 15.05.2025).
7. References & Metrics. RAGas Documentation. E`lektronny`j resurs. URL: https://docs.ragas.io/en/stable/references/metrics/ (data obrashheniya: 15.06.2025).
8. Kenan Agyel. Vector RAG vs Graph RAG vs LightRAG. TDG Global Blog. E`lektronny`j resurs. URL: https://tdg-global.net/ blog/analytics/vector-rag-vs-graph-rag-vs-lightrag/kenan-agyel/ (data obrashheniya: 08.06.2025).
9. Faithfulness. Ragas Documentation. E`lektronny`j resurs. URL: https://docs.ragas.io/en/stable/concepts/metrics/available_metrics/ faithfulness/ (data obrashheniya: 11.06.2025).
10. Advantages and disadvantages of Chatbots: everything you need to know. AIVO. E`lektronny`j resurs. URL: https://www.aivo.co/ blog/advantages-and-disadvantages-of-chatbots (data obrashheniya: 04.06.2025).
11. Jalammar M. Illustrated Transformer. Vizual`noe ob``yasnenie arxitektury` transformerov. Jalammar: obuchayushhie materialy`. URL: https://jalammar.github.io/illustrated-transformer/ (data obrashheniya: 12.06.2025).
12. Jalammar M. Illustrated Word2Vec. Vizual`noe ob``yasnenie arxitektury` Word2Vec. Jalammar: obuchayushhie materialy`. URL: https://jalammar.github.io/illustrated-word2vec/ (data obrashheniya: 12.06.2025).