Datasets:

EdinburghNLP

/

xsum

like 87

Follow

EdinburghNLP - Natural Language Processing Group at the University of Edinburgh 9

I also found some records with empty content but summary

{'document': '', 'summary': "Following the news that Birmingham City Council's children's services are set to be run by a trust after years of problems, BBC News Online answers some common questions.", 'id': '36369226'}

If I use this data set, should I delete the broken(?) until it is corrected?
And how?

https://github.com/EdinburghNLP/XSum/issues/12#issuecomment-617656550

I 'll use dataset from here

Thanks for reporting, @mittu .

We are investigating it.

Please note that first version of this dataset contained only a loading script that used to download the original data from the website of the dataset authors: http://bollin.inf.ed.ac.uk/public/direct/XSUM-EMNLP18-Summary-Data-Original.tar.gz

Once this URL started giving connectivity issues (see #2), we decided to host the original data file (see #3).

We are investigating if there is any issue in the script while processing the data file content, or whether the issues come from the data file content itself.

I think the issue comes from the original data (see below). You should contact the data owners/creators to report it at: https://github.com/EdinburghNLP/XSum

For example, in relation with the example issue reported by @christangttt above, the content of the corresponding data file 36369226.summary is:

[SN]URL[SN]
http://web.archive.org/web/20170119144215/http://www.bbc.co.uk/news/uk-england-birmingham-36369226

[SN]TITLE[SN]
Birmingham City Council: What will happen to children's services?

[SN]FIRST-SENTENCE[SN]
Following the news that Birmingham City Council's children's services are set to be run by a trust after years of problems, BBC News Online answers some common questions.

[SN]RESTBODY[SN]
Share this with
Email
Facebook
Messenger
Messenger
Twitter
Pinterest
WhatsApp
LinkedIn
Copy this link

• the URL and TITLE fields are not used
• the FIRST-SENTENCE field is used as the summary
• the text should come from the RESTBODY field, once removed the lines containing ['Copy this link\n', 'Share this with\n', 'Facebook\n', 'Messenger\n', 'LinkedIn\n', 'Pinterest\n', 'Linkedin\n', 'Email\n', 'WhatsApp\n', 'These are external links and will open in a new window\n', 'Twitter\n']; note there is no text left once removed these sentences.

I also found some data contents is bad , e.g.,34618290.summary
[SN]URL[SN]
http://web.archive.org/web/20170203090905/http://www.bbc.co.uk/news/uk-england-london-34618290

[SN]TITLE[SN]
Bermondsey sailor Albert McKenzie honoured for Zeebrugge raid bravery

[SN]FIRST-SENTENCE[SN]
A statue has been unveiled in honour of the first London sailor to collect the Victoria Cross from King George V.

[SN]RESTBODY[SN]
23 October 2015 Last updated at 18:50 BST
Able Seaman Albert McKenzie took part in the Zeebrugge raid on 23 April 1918.
BBC London reporter Sarah Harris reports.

like this ,the restbody only has updating time and writer.
Perhaps other, more inappropriate data may be included, and it will take time to remove this.

I throw issue.

https://github.com/EdinburghNLP/XSum/issues/29

I am not good at English, so please add your comments if necessary

Thank you for investigating！