M-CTG: Multi Citation Text Generation

Abstract Citation Text Generation (CTG) is crucial in elucidating the relationship between scientific documents. While existing approaches focus primarily on standard summarization or generating single citation texts, they often overlook the complexity of real-world research, where authors frequently summarize multiple studies or discuss relevant information across paragraphs. To address this gap, we present a novel Multi Citation Text Generation (M-CTGP) architecture that leverages prompting knowledge graphs tailored for scientific articles. Our approach incorporates knowledge graphs, source, and target abstracts as input, and introduces the MCG-S2ORC dataset, a curated collection of English language Computer Science academic research papers featuring multiple citation examples. We evaluate our model using three Large Language Models (LLMs), namely Llama, Alpaca, and Vicuna and demonstrate improved performance when incorporating knowledge graphs as prompts for generating citation text. This research showcases the potential of M-CTGP in advancing the field of citation generation, offering an exciting avenue for exploring the intricate relationships between scientific documents.

Status - Pending