On Sun, Apr 16, 2017 at 11:58 PM, John Erling Blad jeblad@gmail.com wrote:
Definitly part of the overall quality. I wonder, do you have any stats om how much positive change the previous attempts have triggered?
[John and I went off-list for me to understand which specific previous attempts he had in mind when asking the above. I have a better sense now and I'm responding to that.]
I'm providing some pointers to indications or controlled experiment results that show how well recommendations in the space of article creation work (Note that I don't have results for the article expansion work to share for now.):
We built an end-to-end system that identifies missing articles in a given language, ranks them according to their importance in that given language, and recommends them to editors who are interested to create them (interest is inferred based on the topic of the articles the editor has edited in the recent past). We ran a controlled experiment and showed that you can increase article creation rate in Wikipedia by a factor of 3.2 if you do personalized recommendations in the setting of the experiment (which was editors receiving recommendations over email) while maintaining the same level of quality as organically created articles on Wikipedia. We also showed that personalized recommendations increase article creation rate by a factor of almost 2 when compared to non-personalized recommendations. If you are interested about the details of this study, you can read the paper that describes it fully at https://arxiv.org/abs/1604.03235. If you prefer a verbal presentation on this topic, I've recently presented this work, why it's important, and some of the work we've started in the article expansion research in CITRIS Exchange seminar series http://citris-uc.org/spring-2017-citris-research-exchange-seminar-series/ in University of California, Berkeley. You can check out the presentation slides https://commons.wikimedia.org/wiki/File:Growing_Wikipedia_Across_Languages_via_Recommendations_CITRIS_20170315.pdf and the video of the presentation https://www.youtube.com/watch?v=lHbnvRwFC_A&index=7&list=PLYTiwx6hV33vqwW7HWyYHMca4H0Ru6KQT .
Outside of the experimental setting, Content Translation is using the recommendation API https://meta.wikimedia.org/wiki/Recommendation_API behind the research I explained above as part of their "Suggestions" feature. Every week, around 180 articles get published on Wikipedia using Suggestions feature alone. This is around 6-9% of all articles created via Content Translation every week. So, we have some evidence that in practice, these recommendations work, too.
I hope this helps.
Best, Leila
Den man. 17. apr. 2017, 02.04 skrev Leila Zia leila@wikimedia.org:
Hi John,
This may be of interest to you:
We are working on building recommendation systems than can help editors identify how to expand already existing articles in Wikipedia. This includes but is not limited to identifying what sections are missing from an article, what citations, what images, infobox information, etc. This
is
research in its early days, if you'd like to follow up with it please
visit
https://meta.wikimedia.org/wiki/Research:Expanding_Wikipedia
_stubs_across_languages
Best, Leila
Leila Zia Senior Research Scientist Wikimedia Foundation
On Sat, Apr 15, 2017 at 2:50 PM, John Erling Blad jeblad@gmail.com wrote:
Are anyone doing any work on automated quality assurance of articles?
Not
the ORES-stuff, that is about creating hints from measured features.
I'm
thinking about verifying existence and completeness of citations, and structure of logical arguments.
John _______________________________________________ Wikimedia-l mailing list, guidelines at: https://meta.wikimedia.org/ wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/ wiki/Wikimedia-l New messages to: Wikimedia-l@lists.wikimedia.org Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l, mailto:wikimedia-l-request@lists.wikimedia.org?subject=unsubscribe
Wikimedia-l mailing list, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l New messages to: Wikimedia-l@lists.wikimedia.org Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l, mailto:wikimedia-l-request@lists.wikimedia.org?subject=unsubscribe
Wikimedia-l mailing list, guidelines at: https://meta.wikimedia.org/wik i/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l New messages to: Wikimedia-l@lists.wikimedia.org Unsubscribe: https://lists.wikimedia.org/mailman/listinfo/wikimedia-l, mailto:wikimedia-l-request@lists.wikimedia.org?subject=unsubscribe