args.me corpus

[janheinrichmerker]

Dataset Information:

The args.me corpus comprises 387 740 arguments. They are crawled from the debate portals Debatewise (14 353 arguments), IDebate.org (13 522 arguments), Debatepedia (21 197 arguments), and Debate.org (338 620 arguments). Moreover, the corpus contains 48 arguments from Canadian Parliament discussions. The arguments are extracted using heuristics that are designed for each debate portal.

Links to Resources:

• Dataset on Zenodo
• Framwework source code

Dataset ID(s):

argsme: Whole corpus

Maybe it might be worth adding entries for the five subsets the corpus was crawled from.

• argsme/debateorg Sub-corpus crawled from Debate.org
• argsme/debatepedia Sub-corpus crawled from Debatepedia
• argsme/debatewise Sub-corpus crawled from Debatewise
• argsme/idebate Sub-corpus crawled from IDebate.org
• argsme/parliamentary Sub-corpus crawled from Canadian Parliament discussions

Supported Entities

• [x] docs
• [ ] queries
• [ ] qrels
• [ ] scoreddocs
• [ ] docpairs

Additional comments/concerns/ideas/etc.

The corpus (or variants of it) is also used in the Touché shared task series:

• 2020
• 2021
• 2022

[seanmacavaney]

Thanks @heinrichreimer! Touché was already on my todo list, so this fits in nicely with that. Currently in progress.

[janheinrichmerker]

I'd be glad to help you with that! Maybe I can even try to implement it by myself and ask you in case of any questions.

[seanmacavaney]

That would be awesome! Let me know if you have any questions.

I think we'll need a third version of the corpus for Touché 2022 because it's based on another version of the corpus using pairs of sentences.