METHOD OF USING ARTIFICIAL INTELLIGENCE AGENTS IN A MULTI-AGENT SYSTEM FOR AUTOMATION OF INTELLIGENT DATA ANALYSIS PROCESSES
DOI:
https://doi.org/10.35546/kntu2078-4481.2024.4.43Keywords:
artificial intelligence, agent, data mining, databases and knowledge, ontological modelAbstract
The article is devoted to the development of a method of using artificial intelligence agents for automation of intelligent data analysis (IAD) using vector and graph databases to accumulate information about the processing pipelines used and search for similar data processing cases and the corresponding chains of application of algorithmic support for the stages of analysis and extraction of models from data. The paper analyzes the possibilities and current state of integration of AI agents with large language models (LLM), which allows significantly expanding the functionality of agents and ensuring the automation of complex IAD processes. The approach proposed in this paper is based on the combined use of the results of ontological modeling of the AI domain, which allows limiting or refining the decisions made by artificial intelligence agents at the relevant stages of the general automated process, and graph databases (Knowledge Graphs) for accumulating knowledge about successful data processing cases. The key results of the research include: development of a methodology for creating multi-agent systems with specialized agents for each stage of the AI process; use of vector databases to search for similar processing cases based on query embeddings; automation of the use of ontological models of the domain as a context for performing tasks by AI agents; iterative approach to data processing with the possibility of improvement based on accumulated experience. The paper presents the results of the structural and functional analysis of the proposed system architecture and a fragment of a graph database for storing knowledge about data processing cases. The advantages and limitations of the use of AI agents are also discussed. The conclusions emphasize the practical value of the proposed approach for improving the efficiency of AI in the face of the growing complexity of processing large amounts of data.
References
Tarasov O., Sahaida P., Podlesny S., Vasylieva L. Categorical-ontological approach to information support of educational activities. Conference on History, Theory and Methodology of Learning: 3rd International (ICHTML 2022), Kryvyi Rih, Ukraine, SHS Web Conf., 2022. Vol. 142. DOI: https://doi.org/10.1051/shsconf/202214203005.
Devansh. An Overview of the AI Agent Landscape. URL : https://medium.com/@machine-learning-made-simple/an-overview-of-the-ai-agent-landscape-9c92009a0a30 (дата звернення 20.11.2024).
Tomaz Bratanic. JSON-based Agents With Ollama & LangChain. URL : https://medium.com/neo4j/json-basedagents-with-ollama-langchain-9cf9ab3c84ef (дата звернення 20.11.2024).
Manpreet Singh. How to Build an AI Agent Army Using ChatGPT: An overview. URL : https://medium.com/ai-advances/how-to-build-an-ai-agent-army-using-chatgpt-a-step-by-step-guide-e60b401c9599 (дата звернення 20.11.2024).
John Mylopoulos. Conceptual Modelling III. Structured Analysis and Design Technique (SADT). URL: http://www.cs.toronto.edu/~jm/2507S/Notes04/SADT.pdf (дата звернення 20.11.2024).
Kenny Vaneetvelde. Atomic Agents in Practice: Creating a Deep, Controllable, Multi-Agent Research System. URL : https://medium.com/generative-ai/atomic-agents-in-practice-creating-a-deep-controllable-multi-agent-researchsystem-e004ad98fec5 (дата звернення 20.11.2024).
Stefan Heindorf, Lukas Blübaum, Nick Düsterhus, Till Werner, Varun Nandkumar Golani, Caglar Demir, and Axel-Cyrille Ngonga Ngomo. Evolearner: Learning description logics with evolutionary algorithms. In Proceedings of the ACM Web Conference 2022, pp. 818-828. 2022.
Satish Kumar. Question Generation AI Agent-Using LangGraph. URL : https://medium.com/@satish-0612/question-generation-ai-agent-using-langgraph-16571ce337f6 (дата звернення 20.11.2024)
Vector database : веб-сайт. URL: https://en.wikipedia.org/wiki/Vector_database (дата звернення 20.11.2024).
Ida Silfverskiöld. Working with Embeddings: Closed versus Open Source. Using techniques to improve semantic search. URL : https://medium.com/towards-data-science/working-with-embeddings-closed-versus-open-source-39491f0b95c2 (дата звернення 20.11.2024).
A free, open-source ontology editor and framework for building intelligent systems : Protégé. URL : https://protege.stanford.edu/ (дата звернення 20.11.2024).
Anthony Alcaraz. Orchestrating a Swarm of AI Agents to Accomplish Complex Goals : A theorical approach. URL : https://medium.com/ai-in-plain-english/orchestrating-a-swarm-of-ai-agents-to-accomplish-complex-goals-atheorical-approach-57241b614b46 (дата звернення 20.11.2024).
