MindGraph.Ai: Knowledge Graphs Creation using Large Language Models

Large Language Models (LLMs) like GPT-4, built upon the Transformer architecture [1], have revolutionized natural language processing but often lack direct access to structured factual knowledge [2]." Knowledge graphs (KGs) offer structured data but struggle with scalability, evolution, and updates. Integrating LLM with KG enhances knowledge retrieval, reasoning, and adaptability. This research proposes a question-answering system using Retrieved Generation (RAG) and Neo4j to bridge LLM and KG. The system refines entity extraction, relationship mapping, and dynamic KG updates, improving contextual relevance and precision. Performance evaluations show that GPT-4 outperforms other models, achieving 90% accuracy and 88% F1 score, demonstrating strong language understanding and relationship extraction, although with 85% specificity. Advanced node filtering and query interpretation improve accuracy, particularly in complex domains such as legal documents. A user-friendly interface facilitates efficient node access and data export, ensuring seamless knowledge utilization. The proposed system advances AI-driven knowledge retrieval by leveraging GPT-4’s adaptability and KGs’ structured knowledge, making question-answering tasks more reliable, scalable, and context-aware. This approach significantly improves the usability and effectiveness of AI systems in handling complex queries across various domains.