BioGen: Generating Biography Summary under Table Guidance on Wikipedia

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4752–4757 August 1–6, 2021. ©2021 Association for Computational Linguistics 4752 BioGen: Generating Biography Summary under Table Guidance on Wikipedia Shen Gao ♠∗, Xiuying Chen ♠♣∗, Chang Liu ♠♣, Dongyan Zhao ♠♣♥†, Rui Yan ♦ ♠Wangxuan Institute of Computer Technology, Peking University, Beijing, China ♣Center for Data Science, Peking University, Beijing, China ♦Gaoling School of Artificial Intelligence, Renmin University of China, Beijing, China ♥State Key Laboratory of Media Convergence Production Technology and Systems {shengao, xy-chen, liuchang97, zhaody}@pku.edu.cn ruiyan@ruc.edu.cn Abstract Capturing the salient information from an in- put article has been a long-standing challenge for summarization. On Wikipedia, most of the wiki pages about people contain a factual ta- ble that lists the basic properties of the peo- ple. Illuminatingly, a factual table can be re- garded as a natural summary of the key in- formation in the corresponding article. Thus, in this paper we propose the task of table- guided abstractive biography summarization, which utilizes factual tables to capture im- portant information and then generate a sum- mary of a biography. We first introduce the TaGS (Table-Guided Summarization) dataset1, the first large-scale biography summarization dataset with tables. Next, we report some statistics about this dataset to validate the qual- ity of the dataset. We also benchmark sev- eral commonly used summarization methods on TaGS and hope this will inspire more excit- ing methods. 1 Introduction Text summarization generates a short text version of a long passage which retains the most important information. Recently, two kinds of approaches have been proposed for automatic text summariza- tion. One is extractive summarization (Nallapati et al., 2017; Liu and Lapata, 2019), which directly selects salient sentences from the passage to create a summary. The other is abstractive summariza- tion (See et al., 2017; Hsu et al., 2018a), which aims to concisely paraphrase the input article. In both methods, the summary should always focus on important information, though a document may include trivial facts. ∗Equal contribution. Ordering is decided by a coin flip. †Corresponding Author: Dongyan Zhao 1https://github.com/gsh199449/ table-summ To focus on the main information when gen- erating summaries, some researchers propose to incorporate manifold information to improve the performance. Narayan et al. (2017) proposed to incorporate the figures and Gao et al. (2019b) in- vestigated the using of reader comments for more effective summarization. As another type of side information, factual tables provide a natural sum- mary of the biography document. On Wikipedia, in each wiki page about people, there is a factual table (infobox) on the right side of the page sum- marizing the main properties. Clearly, infobox is helpful for capturing the salient information during summarizing the biography. However, no existing work takes advantage of tables, though are widely available in the biography on Wikipedia. In this paper, we propose Table-Guided Summa- rization (TaGS) dataset, the first large-scale biog- raphy summarization dataset with tables. And we report some statistics and three important charac- teristics of this dataset to verify its quality. The first …

Content truncated to 500 words