The paper "Knowledge Graph-Enhanced Large Language Models via Path Selection" addresses the issue of hallucination in Large Language Models (LLMs) by incorporating external knowledge from Knowledge Graphs (KGs). The authors propose a framework called KELP (Knowledge Graph-Enhanced Large Language Models via Path Selection) to improve the factual accuracy of LLM outputs. KELP consists of three stages: knowledge path extraction, sample encoding, and fine-grained path selection. During knowledge path extraction, the model identifies relevant paths in the KG based on entities in the input text. Sample encoding involves training a path-text encoder to measure the similarity between input texts and extracted paths, capturing potentially impactful knowledge. Fine-grained path selection uses coverage rules to refine the selected paths, ensuring flexibility and capturing diverse and representative knowledge. The effectiveness of KELP is validated through experiments on real-world datasets, demonstrating its ability to enhance LLM performance by incorporating external knowledge effectively.The paper "Knowledge Graph-Enhanced Large Language Models via Path Selection" addresses the issue of hallucination in Large Language Models (LLMs) by incorporating external knowledge from Knowledge Graphs (KGs). The authors propose a framework called KELP (Knowledge Graph-Enhanced Large Language Models via Path Selection) to improve the factual accuracy of LLM outputs. KELP consists of three stages: knowledge path extraction, sample encoding, and fine-grained path selection. During knowledge path extraction, the model identifies relevant paths in the KG based on entities in the input text. Sample encoding involves training a path-text encoder to measure the similarity between input texts and extracted paths, capturing potentially impactful knowledge. Fine-grained path selection uses coverage rules to refine the selected paths, ensuring flexibility and capturing diverse and representative knowledge. The effectiveness of KELP is validated through experiments on real-world datasets, demonstrating its ability to enhance LLM performance by incorporating external knowledge effectively.