Enhancing Scientific Papers Summarization with Citation Graph

Previous work for text summarization in scientific domain mainly focused on the content of input document, but seldom considering its citation network. However, papers are full uncommon domain-specific terms, making it almost impossible model to understand true meaning without help relevant research community. In this paper, we redefine task by utilizing their graph and propose a graph-based CGSum which can incorporate information both source paper references. addition, construct novel dataset Semantic Scholar Network (SSN) contains 141K different domains 661K relationships. The entire constitutes large connected graph. Extensive experiments show that our achieve competitive performance when compared with pretrained models even simple architecture. results also indicates is crucial better generate high-quality summaries.