Automatic Generation of Related Work Sections in Scientific Papers: An Optimization Approach

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1624–1633, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Automatic Generation of Related Work Sections in Scientific Papers: An Optimization Approach Yue Hu and Xiaojun Wan Institute of Computer Science and Technology The MOE Key Laboratory of Computational Linguistics Peking University, Beijing, China {ayue.hu,wanxiaojun}@pku.edu.cn Abstract In this paper, we investigate a challeng- ing task of automatic related work gener- ation. Given multiple reference papers as input, the task aims to generate a related work section for a target paper. The gen- erated related work section can be used as a draft for the author to complete his or her final related work section. We propose our Automatic Related Work Generation system called ARWG to ad- dress this task. It first exploits a PLSA model to split the sentence set of the giv- en papers into different topic-biased parts, and then applies regression models to learn the importance of the sentences. At last it employs an optimization frame- work to generate the related work section. Our evaluation results on a test set of 150 target papers along with their reference papers show that our proposed ARWG system can generate related work sec- tions with better quality. A user study is also performed to show ARWG can achieve an improvement over generic multi-document summarization baselines. 1 Introduction The related work section is an important part of a paper. An author often needs to help readers to understand the context of his or her research problem and compare his or her current work with previous works. A related work section is often used for this purpose to show the differ- ences and advantages of his or her work, com- pared with related research works. In this study, we attempt to automatically generate a related work section for a target academic paper with its reference papers. This kind of related work sec- tions can be used as a basis to reduce the author’s time and effort when he or she wants to complete his or her final related work section. Automatic related work section generation is a very challenging task. It can be considered a top- ic-biased, multiple-document summarization problem. The input is a target academic paper, which has no related work section, along with its reference papers. The goal is to create a related work section that describes the related works and addresses the relationship between the target pa- per and the reference papers. Here we assume that the set of reference papers has been given as part of the input. Existing works in the NLP and recommendation systems communities have al- ready focused on the task of finding reference papers. For example, citation prediction (Nal- lapati et al., 2008) aims at finding individual pa- per citation patterns. Generally speaking, automatic related work section generation is a strikingly different prob- lem and it is much more difficult in comparison with general multi-document summarization tasks. For example, multi-document …

Content truncated to 500 words