Bringing Two Worlds Together: The Convergence of Large Language Models and Knowledge Graphs

Large Language Models (LLMs) have transformed numerous fields with their strong capabilities in generalization and inference, yet they face challenges such as dependence on probabilistic reasoning, lack of factual knowledge, and limited interpretability. Knowledge Graphs (KGs), with their structured factual information, can address these gaps by providing external knowledge, though they also struggle with issues like scalability and representing new or unseen facts. Integrating LLMs and KGs offers a complementary solution, as both can mutually enhance each other. This article categorizes and reviews three main frameworks: (1) LLM-enhanced KGs, where LLMs support KG tasks like embedding, completion, and question answering; (2) KGenhanced LLMs, which integrate KGs to improve LLMs' factual understanding; and (3) Collaborative KGs and LLMs, where both systems interact symbiotically, enabling bidirectional reasoning. We review representative methods, analyze current limitations, and discuss future research directions and applications.