Scientific Paper Extractive Summarization Enhanced by Citation Graphs

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 4053 - 4062 December 7-11, 2022 ©2022 Association for Computational Linguistics Scientific Paper Extractive Summarization Enhanced by Citation Graphs Xiuying Chen1∗, Mingzhe Li2∗, Shen Gao3, Rui Yan4, Xin Gao1†, Xiangliang Zhang5,1† 1 Computational Bioscience Reseach Center, KAUST 2 Ant Group 3 School of Computer Science and Technology, Shandong University 4 Gaoling School of Artificial Intelligence, Renmin University of China 5 University of Notre Dame xiuying.chen@kaust.edu.sa, limingzhe.lmz@antgroup.com Abstract In a citation graph, adjacent paper nodes share related scientific terms and topics. The graph thus conveys unique structure information of document-level relatedness that can be utilized in the paper summarization task, for exploring beyond the intra-document information. In this work, we focus on leveraging citation graphs to improve scientific paper extractive summariza- tion under different settings. We first propose a Multi-granularity Unsupervised Summariza- tion model (MUS) as a simple and low-cost so- lution to the task. MUS finetunes a pre-trained encoder model on the citation graph by link prediction tasks. Then, the abstract sentences are extracted from the corresponding paper con- sidering multi-granularity information. Prelim- inary results demonstrate that citation graph is helpful even in a simple unsupervised frame- work. Motivated by this, we next propose a Graph-based Supervised Summarization model (GSS) to achieve more accurate results on the task when large-scale labeled data are available. Apart from employing the link prediction as an auxiliary task, GSS introduces a gated sentence encoder and a graph information fusion mod- ule to take advantage of the graph information to polish the sentence representation. Experi- ments on a public benchmark dataset show that MUS and GSS bring substantial improvements over the prior state-of-the-art model. 1 Introduction Text summarization is to automatically glean the most important concepts from an article, removing secondary or redundant concepts. Among various summarization tasks such as news summarization (Wang et al., 2020), dialog summarization (Zhang et al., 2021), and timeline summarization (Chen et al., 2019), scientific paper summarization re- mains a challenging task, since scientific papers are usually longer, and full of complex concepts * Equal contribution. † Corresponding author. Target paper abstract: ...we propose a multi-granularity interaction network for extractive and abstractive multi-document summarization... We employ attention mechanisms to interact between different granularity of semantic representations... Our proposed model outperforms all strong baseline methods... Paper 1 abstract: ...We propose a unified model combining the strength of extractive and abstractive summarization... Paper 2 abstract: ...we propose HIBERT (HIerachical Bidirectional Encoder Representations from Transformers)... Paper 3 abstract: ... Our empirical analysis shows state-of-the-art performance on several multi document datasets. Paper 4 abstract: ...The Hidden Markov Model (HMM) for part-of-speech (POS) tagging is typically based on tag trigrams... Paper 5 abstract: ...Paraphrase recognition is used in a number of applications such as tutoring systems... ... Target paper Paper 1 Paper 3 Paper 2 Paper 4 Paper 5 Figure 1: A small research community on the subject of text summarization. Arrows denote citation relation- ships. Sentences with shared domain-specific terms are …

Content truncated to 500 words