This looks really cool and valuable. Thanks for your work on it, Aiko!
I took your prototype for a test run, but was a little surprised by the
first three tasks it gave me:
For all three of them, the text appears to already be referenced. (For the
second one, the second sentence doesn't have the reference immediately
following it, so I can see why that would be a problem, but the tool
highlighted the first sentence as well.) Is this a bug, or is the tool
telling me that the sources used are unreliable, or am I just
On Sat, Mar 7, 2020 at 9:03 AM Ai-Jou Chou <qwanqwanro(a)gmail.com> wrote:
I’m happy to announce the outcome of an Outreachy internship
<https://phabricator.wikimedia.org/T233707> that I’m finishing up. It is a
new tool and public dataset named Citation Detective which tool developers
and researchers can now use for their projects.
Citation Detective <https://meta.wikimedia.org/wiki/Citation_Detective>
contains sentences that have been identified as needing a citation using a
machine learning-based classifier published earlier last year
<https://arxiv.org/pdf/1902.11116.pdf> by WMF researchers and
collaborators. As part of Outreachy, I developed a tool
<https://github.com/AikoChou/citationdetective> (hosted on Toolforge
<https://tools.wmflabs.org>) to run through Wikipedia and extract
high-scoring sentences along with contextual information.
As an example use case for this data, I also created a proof of concept for
integrating Citation Detective and Citation Hunt
<https://tools.wmflabs.org/citationhunt>. Check out my prototype Citation
Hunt <https://tools.wmflabs.org/aiko-citationhunt>, which uses Citation
Detective to import sentences that would not normally be featured in
Citation Hunt. The repository for that is here
This dataset currently includes sentences from ~120,000 randomly selected
articles from the English Wikipedia. In future work, we hope to expand this
to more language Wikipedia projects and a greater number of articles. It is
also possible to expand the database to contain more fields in a future
version according to feedback from tool developers and researchers. More
use cases for this type of data were identified in a design research
conducted last year by Jonathan Morgan.
You can find more information in our Wiki Workshop submission
and in my blog
<https://rollingmist.home.blog/> which documented the whole
Thank you very much!
Wikitech-l mailing list