KnoBuilder: An LLM-Agent for Autonomous and Personalized Knowledge Graph Construction from Unstructured Text

This paper introduces KnoBuilder, a novel LLM-based agentic framework for autonomous construction of personalized knowledge graphs from unstructured text corpora. Addressing the limitations of traditional knowledge graph construction methods and one-shot LLM extraction approaches, KnoBuilder implements a synergistic loop between an LLM agent and a dynamically evolving knowledge graph. The framework features strategic planning for knowledge acquisition, self-refining information extraction with multi-stage validation, and dynamic consolidation maintaining graph coherence. Extensive evaluation on scientific corpora demonstrates that KnoBuilder significantly outperforms state-of-the-art baselines, achieving 85% F1-score in extraction quality, 46% improvement in acquisition efficiency, 91% entity resolution accuracy, and superior performance in complex query answering, while maintaining coherent graph structures with 96% consistency scores.