This paper addresses the challenge of generating multi-sentence citations for scientific documents, which is crucial for understanding the connections between papers. The authors propose a method that leverages Large Language Models (LLMs) to generate coherent paragraphs containing multiple citations. They introduce a curated dataset named MCG-S2ORC, which includes English-language academic research papers in Computer Science, featuring multiple citation instances. The study evaluates three LLMs—LLaMA, Alpaca, and Vicuna—to determine the most effective model for this task. Additionally, the authors demonstrate that integrating knowledge graphs from target papers into the prompts significantly enhances the performance of the models. The research highlights the potential of LLMs in citation generation, offering a promising avenue for exploring the intricate connections between scientific documents. The contributions of the paper include a Citation Generation architecture, the creation of the MCG-S2ORC dataset, and the demonstration of the importance of incorporating knowledge graphs in the prompt structure. The experiments show that Vicuna outperforms the other models, and the integration of knowledge graphs further improves the quality of the generated citations. The paper concludes by discussing the limitations and future research directions, emphasizing the potential of knowledge graphs in enhancing citation generation.This paper addresses the challenge of generating multi-sentence citations for scientific documents, which is crucial for understanding the connections between papers. The authors propose a method that leverages Large Language Models (LLMs) to generate coherent paragraphs containing multiple citations. They introduce a curated dataset named MCG-S2ORC, which includes English-language academic research papers in Computer Science, featuring multiple citation instances. The study evaluates three LLMs—LLaMA, Alpaca, and Vicuna—to determine the most effective model for this task. Additionally, the authors demonstrate that integrating knowledge graphs from target papers into the prompts significantly enhances the performance of the models. The research highlights the potential of LLMs in citation generation, offering a promising avenue for exploring the intricate connections between scientific documents. The contributions of the paper include a Citation Generation architecture, the creation of the MCG-S2ORC dataset, and the demonstration of the importance of incorporating knowledge graphs in the prompt structure. The experiments show that Vicuna outperforms the other models, and the integration of knowledge graphs further improves the quality of the generated citations. The paper concludes by discussing the limitations and future research directions, emphasizing the potential of knowledge graphs in enhancing citation generation.