Hybrid Multi-Agent Systems and Graph Analytics for Enhanced Knowledge Graph Generation from Unstructured Data

Author : Cristian Ulmeanu, Filipoiu Ciprian Vicențiu

Abstract :Unstructured textual data, such as transcripts, contain rich yet challenging-to-use information for analytics. We introduce a generalizable framework to construct semantically rich knowledge graphs (KGs) from unstructured text, combining an agentic extraction architecture with programmatic graph refinement. Our approach employs specialized agents to iteratively extract and organize content into nodes and relations, while a programmatic schema layer explicitly incorporates contextual metadata as relational keys. To enhance semantic density and reduce redundancy, we apply Louvain community detection coupled with agent driven consolidation. Using meeting transcripts as an experiment, we validate our approach’s generalizability to diverse domains. Preliminary evaluations indicate that our refined KGs notably improve downstream tasks, such as query answering by large language models, compared to raw text alone. Thus, our methodology effectively bridges unstructured text and structured knowledge, providing a scalable pathway for enriching large language model performance.

Keywords :Knowledge Graph Construction of Unstructured Text, Natural Language Processing (NLP), Multi-Agent Systems, Graph Clustering, Louvain Algorithm, Decision Support.

Conference Name :International Conference on Emerging Trends in Artificial Intelligence, Data Science and Cloud Computing (ICETADSCC-25)

Conference Place Dubai,UAE

Conference Date 20th Nov 2025