Comparative Graph-based Summarization of Scientific Papers Guided by Comparative Citations

Proceedings of the 29th International Conference on Computational Linguistics, pages 5978–5988 October 12–17, 2022. 5978 Comparative Graph-based Summarization of Scientific Papers Guided by Comparative Citations Jingqiang Chen1,†,∗, Chaoxiang Cai1,†, Xiaorui Jiang2, Kejia Chen1 1School of Computer Science, Nanjing University of Posts and Telecommunications, China 2Centre for Computational Science and Mathematical Modelling, Coventry University, UK {cjq,1220044905,chenkj}@njupt.edu.cn, xiaorui.jiang@coventry.ac.uk Abstract With the rapid growth of scientific papers, un- derstanding the changes and trends in a re- search area is rather time-consuming. The first challenge is to find related and comparable ar- ticles for the research. Comparative citations compare co-cited papers in a citation sentence and can serve as good guidance for researchers to track a research area. We thus go through comparative citations to find comparable ob- jects and build a comparative scientific summa- rization corpus (CSSC). And then, we propose the comparative graph-based summarization (CGSUM) method to create comparative sum- maries using citations as guidance. The com- parative graph is constructed using sentences as nodes and three different relationships of sen- tences as edges. The relationship that sentences occur in the same paper is used to calculate the salience of sentences, the relationship that sen- tences occur in two different papers is used to calculate the difference between sentences, and the relationship that sentences are related to ci- tations is used to calculate the commonality of sentences. Experiments show that CGSUM out- performs comparative baselines on CSSC and performs well on DUC2006 and DUC2007. 1 Introduction Today, the transient and rapidly evolving research areas and the numerous published research articles require researchers to orient themselves and dis- cover the changes of the research area (Marrone, 2020). In order to reduce the burden of researchers, a solution is to find and compare related articles in the research area, and automatically create com- parative summaries showing commonalities and differences of the articles where differences mean changes. The first problem is how to find related and comparable articles, and the second problem is how to create comparative summaries. † The authors have contributed equally to this work. ∗ Jingqiang Chen is the corresponding author. Comparative citation from BERT (Devlin et al., 2018): Unlike recent language representation models (Peters et al., 2018; Radford et al., 2018), BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. [Devlin et al., 2018]; [Peters et al., 2018; Radford et al., 2018]; language representation models. Citing paper: Cited papers: Comparable topic: ① We introduce a new type of deep contextualized word representation ... use vectors derived from a bidirectional LSTM ... [Peters et al., 2018] ② We introduced a framework for achieving strong natural language understanding ... pre-training on a diverse corpus with long stretches of contiguous text ... [Radford et al., 2018] ③ We introduce a new language representation model ... uses masked language models ... [Devlin et al., 2018] Comparative summary based on the comparative citation: Figure 1: An example from our dataset. The compar- …

Content truncated to 500 words