Multi-agent system model for knowledge graph construction
DOI:
https://doi.org/10.34121/1028-9763-2026-2-79-91Keywords:
кnowledge graphs, LLM, multi-agent systems, GrafRAGAbstract
Currently, the use of knowledge graphs is a powerful way to represent information in many areas of the economy, such as cartography, construction planning, logistics, biology, medicine, etc. However, building a knowledge graph from unstructured text is rather difficult. It requires identifying entities and their relationships, writing manual extraction rules, or using specialized machine learning models. At the same time, large language models (LLM), which are currently developing very rapidly, can be successfully used to solve this problem. The use of LLM, when reinforced by the simultaneous use of multi-agent systems, should result in a significant increase in the stability and adaptability of knowledge graphs. The paper investigates and provides examples of the transformation of the model of relational relationships between tables in a structured database to a model of connections between graph nodes and examples of the work and interaction of agents in a multi-agent architecture. The list and features of the main frameworks for agent-based LLMs in 2025 are presented. A model of a multi-agent knowledge graph construction system using LLM is studied and created. The work of agents of structured and unstructured model data is described. The model uses the technology of generation supplemented by search — RAG (Retrieval-Augmented Generation). The GrafRAG agent is responsible for working with unstructured text. The creation of plans for building a knowledge graph from structured and extracting knowledge from unstructured texts is shown. Examples of functions for loading nodes and connections between them from CSV files in the Cypher language into the Neo4j database are given. A diagram of extracting entities and relations from text when building a knowledge graph is constructed. The work can be useful for building multi-agent knowledge graph systems that use LLM, since their mutual work leads to a reduction in errors and hallucinations when responding to user requests. Таbl.: 1. Figs.: 7. Refs.: 19 titles.
References
1. SAP. Що таке велика мовна модель (LLM) — SAP. URL: https://www.sap.com/ukraine/resources/what-is-large-language-model.
2. Software in the Age of AI. URL: https://www.latent.space/p/s3.
3. The platform where the machine learning community collaborates on models, datasets, and applications. URL: https://huggingface.co/models.
4. IBM’s Granite foundation models are targeted for business. URL: https://www.ibm.com/think/news/granite-foundation-models.
5. What is knowledge graph — IBM. URL: https://www.ibm.com/think/topics/knowledge-graph.
6. What is knowledge graph — neo4j. URL: https://neo4j.com/blog/knowledge-graph/what-is-knowledge-graph/.
7. Що таке мультиагентні системи — SAP. URL: https://www.sap.com/ukraine/resources/what-are-multi-agent-systems.
8. Мультиагентні системи — MAS — InterNetri. URL: https://internetri.net/qntm/2025/09/23/multyagentni-systemy-mas/.
9. How to Build Knowledge Graphs using LLMs on Local Machine. URL: https://medium.com/the-muse-junction/how-to-build-knowledge-graphs-using-llms-on-local-machines-5926261c04eb.
10. Neo4j Graph Database. URL: https://neo4j.com/docs.
11. Building a Knowledge Graph From Scratch Using LLMs. URL: https://towardsdatascience.com/building-a-knowledge-graph-from-scratch-using-llms-f6f677a17f07/.
12. Building Knowledge Graphs Using Large Language Models. URL: https://medium.com/@shuchawl/building-knowledge-graphs-using-large-language-models-07da1935b21a.
13. neo4j-contrib/agentic-kg. URL: https://github.com/neo4j-contrib/agentic-kg.
14.Multi-agent LLMs in 2025 [+frameworks]. URL: https://www.superannotate.com/blog/multi-agent-llms.
15. Choosing the Right LLM Agent Framework in 2025. URL: https://botpress.com/blog/llm-agent-framework.
16. Top 5 multi-agent frameworks for Building Automation Systems. URL: https://www.zams.com/blog/multi-agent-frameworks.
17. Тop 9 AI Agent Frameworks as of September 2025. URL: https://www.shakudo.io/blog/top-9-ai-agent-frameworks.
18. Оptimizing the Interface Between Knowledge Graphs and LLMs for Complex Reasoning. URL: https://arxiv.org/html/2505.24478v1.
19. Automate building knowledge graphs with LLMs and Neo4J. URL: https://blog.gopenai.com/automate-building-knowledge-graphs-from-scientific-abstracts-with-llms-and-neo4j-2-ways-b9d5152a12da.
232
74
