KRAGEN: a knowledge graph-enhanced RAG framework for biomedical problem solving using large language models

Abstract

In this article, we introduced KRAGEN that combines the reasoning power of LLMs and the factual feedback of knowledge graphs to answer complex questions and problems. KRAGEN uses an advanced prompting technique called graph of thoughts to model and execute the reasoning process of the language model, and provides a visual interface that shows the logic and evidence behind the generated responses. We demonstrated the features and benefits of KRAGEN using a real-world example of an Alzheimer’s disease knowledge graph, AlzKB. We showed how KRAGEN can retrieve relevant facts from the knowledge gra