Multi-News: A Large-Scale Multi-Document Summarization Dataset and Abstractive Hierarchical Model

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1074–1084 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 1074 Multi-News: a Large-Scale Multi-Document Summarization Dataset and Abstractive Hierarchical Model Alexander R. Fabbri Irene Li Tianwei She Suyi Li Dragomir R. Radev Department of Computer Science Yale University {alexander.fabbri,irene.li,tianwei.she,suyi.li,dragomir.radev}@yale.edu Abstract Automatic generation of summaries from mul- tiple news articles is a valuable tool as the number of online publications grows rapidly. Single document summarization (SDS) sys- tems have benefited from advances in neu- ral encoder-decoder model thanks to the avail- ability of large datasets. However, multi- document summarization (MDS) of news ar- ticles has been limited to datasets of a couple of hundred examples. In this paper, we in- troduce Multi-News, the first large-scale MDS news dataset. Additionally, we propose an end-to-end model which incorporates a tradi- tional extractive summarization model with a standard SDS model and achieves competitive results on MDS datasets. We benchmark sev- eral methods on Multi-News and release our data and code in hope that this work will pro- mote advances in summarization in the multi- document setting1. 1 Introduction Summarization is a central problem in Natural Language Processing with increasing applications as the desire to receive content in a concise and easily-understood format increases. Recent ad- vances in neural methods for text summariza- tion have largely been applied in the setting of single-document news summarization and head- line generation (Rush et al., 2015; See et al., 2017; Gehrmann et al., 2018). These works take advan- tage of large datasets such as the Gigaword Corpus (Napoles et al., 2012), the CNN/Daily Mail (CN- NDM) dataset (Hermann et al., 2015), the New York Times dataset (NYT, 2008) and the News- room corpus (Grusky et al., 2018), which con- tain on the order of hundreds of thousands to mil- lions of article-summary pairs. However, multi- document summarization (MDS), which aims to 1https://github.com/Alex-Fabbri/ Multi-News Source 1 Meng Wanzhou, Huawei’s chief financial officer and deputy chair, was arrested in Vancouver on 1 December. Details of the arrest have not been released... Source 2 A Chinese foreign ministry spokesman said on Thurs- day that Beijing had separately called on the US and Canada to “clarify the reasons for the detention ”imme- diately and “immediately release the detained person ”. The spokesman... Source 3 Canadian officials have arrested Meng Wanzhou, the chief financial officer and deputy chair of the board for the Chi- nese tech giant Huawei,...Meng was arrested in Vancou- ver on Saturday and is being sought for extradition by the United States. A bail hearing has been set for Friday... Summary ...Canadian authorities say she was being sought for extra- dition to the US, where the company is being investigated for possible violation of sanctions against Iran. Canada’s justice department said Meng was arrested in Vancouver on Dec. 1... China’s embassy in Ottawa released a state- ment.. “The Chinese side has lodged stern representations with the US and Canadian side, and urged …

Content truncated to 500 words