Lankide eztabaida:EpochFail

Ongi etorri, EpochFail, euskarazko Wikipediara! Hau zure eztabaida orria da, beste wikilarien mezuak jasotzeko balioko dizuna. Gainera, lankide orria ere baduzu; nahi izanez gero, hor jar ditzakezu zeure datuak, interesak, hitz egiten dituzun hizkuntzak... (zeure lankide orria egiteko informazioa nahi baduzu, ikus orrialde hau). Nahi baduzu ere zure proba orrian idazten has zaitezke. Lagungarriak izan daitezkeen loturak Wikipedian aritzeko bost euskarriak ezagutu Ezagutu Wikipediako politikak Estilo gida ezagutu Hasi orain! Pausoz pauso aurrera Wikitestu tutoriala Aldaketak fin-fin kodeak erabiliz Orokorrak Praktikoak Teknikoak Irakasleak Wikipedian laguntzeko modu ugari daude. Horietako batzuk Wikipedian nola lagundu dezakezun argitzen duen orrian bildu ditugu. Artikulu jakin batez ohar edo iradokizunen bat egiteko, artikulu horren eztabaida orrian idatz dezakezu. Wikipediari buruzko galdera, ohar edo iradokizun orokorra baldin bada, berriz, Wikipediako Txokoa duzu horretarako toki egokiena. Eta hainbat Wikiproiektu ditugu, gai jakin batzuetan parte hartzen lagunduko dizutenak. Wikilari jakin bati idazteko edo erantzuteko, haren eztabaida orrian idatzi beharko duzu. Bestela, ez zaio mezu bat duela jakinarazteko oharra azalduko, eta agian ez da konturatuko idatzi diozula. Eztabaida orriaren goialdean «Gehitu atala» lotura aurkituko duzu. Hor klik eginez, mezua sortzeko aukera izango duzu. Mesedez, ez zaitez ahaztu mezuaren bukaeran zure erabiltzaile sinadura edo zigilua ezartzeaz; bestela, ezingo dugu jakin nork idatzi duen, nori erantzun behar diogun. ES: Si no comprende o no sabe escribir en nuestra lengua y desea comunicarse con nosotras y nosotros, por favor, acuda a la Embajada. FR: Si vous ne comprenez pas ou ne savez pas écrire dans notre langue et souhaitez communiquer avec nous, s'il vous plaît, visitez l'Ambassade. EN: If you do not understand or you cannot write in our language and you want to tell us something, please visit the Embassy. DE: Wenn Sie unsere Sprache nicht verstehen oder nicht schreiben können und Sie bei uns mitreden möchten, dann besuchen Sie bitte die Botschaft. RU: Если Вы не понимаете или не можете писать на нашем языке, но хотите поделиться с нами чем-либо, пожалуйста, посетите наше посольство.

--EnzaiBot (eztabaida) 11:26, 16 uztaila 2015 (UTC)

Enable "articlequality" system for everyone

Hello! "Articlequality" system is working fine, and it would be useful to have it included as an opt-out gadget for everyone. How do you see it? Thanks!-Theklan · · 13:36, 15 urria 2018 (UTC)

This is a good idea. It's been on my radar. Thanks for pushing on this. FWIW I just fixed an issue with the gadget so it'll continue working.

Either you could make me an admin on euwiki and I could try to make this work or you could take the following steps and they could work.

1. Create the page: MediaWiki:Gadget-ArticleQuality.js. And add the following content to it.
   • mw.loader.load( '//eu.wikipedia.org/wiki/Lankide:EpochFail/ArticleQuality.js&action=raw&ctype=text/javascript' );
2. Then on MediaWiki:Gadgets-definition, add the following line to an appropriate section:
   • ArticleQuality[ResourceLoader|default]|ArticleQuality.js Translate the first "ArticleQuality" to something more reasonable. That will appear in Special:Preferences

After you have done that, the option should be enabled by default in Special:Preferences for logged-in users. --EpochFail (eztabaida) 01:24, 24 urria 2018 (UTC)

Ping Theklan. :) I've now updated the docs at en:User:EpochFail/ArticleQuality so you could work from those. --EpochFail (eztabaida) 16:58, 13 azaroa 2018 (UTC)

Thanks! I've done the work (by the way, it's Gadget in singular) but it doesn't seem to be working. Maybe the gadget needs loading something else? -Theklan · · 18:51, 13 azaroa 2018 (UTC)

Aha! I had a typo in the javascript. Change the following line:

mw.loader.load( '//eu.wikipedia.org/wiki/Lankide:EpochFail/ArticleQuality.js&action=raw&ctype=text/javascript' ); to:

mw.loader.load( '//eu.wikipedia.org/w/index.php?title=Lankide:EpochFail/ArticleQuality.js&action=raw&ctype=text/javascript' );

--EpochFail (eztabaida) 19:38, 13 azaroa 2018 (UTC)

thanks! -Theklan · · 19:56, 13 azaroa 2018 (UTC)

Minimun and maximum score

Hello! I have noticed that the minimum score is currently 1,79 and the maximum seems to be 5,5. People here is mostly used to something from 0 to 10... would it be possible to make a recalibration or is it completely impossible? -Theklan · · 16:30, 22 azaroa 2018 (UTC)

Hey! Good question. It's possible. Can you file a task on phabricator and I'll get to it? --EpochFail (eztabaida) 14:11, 24 urtarrila 2019 (UTC)

Forgot to ping Theklan. :) --EpochFail (eztabaida) 14:11, 24 urtarrila 2019 (UTC)

Was easier than I thought. Txantiloi:Done Sorry for the delay. I'd somehow missed the cross-wiki notification for the original request. --EpochFail (eztabaida) 14:16, 24 urtarrila 2019 (UTC)

Thanks! I filled it in Phabricator but I see it's done. Is it from 0 to 10? Our "best" article is Lehen Mundu Gerra and it doesn't show more than a 9. -Theklan · · 20:58, 24 urtarrila 2019 (UTC)

Theklan, Hmm. It *should* max out at a 10. I see you made some edits to push the value around a bit. When I ask ORES to score the most recent version of that article, I get the following:

            "probability": {
              "B": 0.060913065596473664,
              "C": 0.03097760657241707,
              "FA": 0.6898985627430138,
              "GA": 0.19219889254871875,
              "Start": 0.01392400082011055,
              "Stub": 0.012087871719266383
            }

The model is pretty confident that the page is top quality ("FA"), but there's a small amount of probability that it might be second-best quality ("GA"). When I scaled the scoring system so that it tops out at 10, "FA" was worth 10 and "GA" was worth 8. So we land somewhere inbetween at 8.95. One good way for us to adjust the way that the scores are scaled is to make the scale more closely match the way we think about quality. E.g. if "GA" is closer to "FA" than "B" is to "GA" we might give "GA" a value of 9. If we were to do that, then that would pull the weighted sum up to 9.14.

Essentially, I think the reason why you aren't seeing scores close to 10 is because of the model's lack of confidence. I think it would be very unlikely to see ORES give more than 70% weight to FA even for an article that is clearly FA. This means that the model is *right* in that it chooses the most likely quality class correctly, but not sure. Short of pushing the weighting around to give the impression of higher confidence, we can always train ORES better. In this case, I expect that labeling more articles could have a great effect. But that's a lot of work. What do you think? Should we just fudge the scale a little bit or would you be interested in doing the work to get more training examples for ORES? I can get Wikilabels set up again with new samples for labeling. --EpochFail (eztabaida) 19:28, 30 urtarrila 2019 (UTC)

It would be interesting to have a new scoring round, as we have changed lot of articles, now lists are not considered articles so they would be excluded and we have a good bunch of FA around. Also, we labelled as C all the automatically created articles about french comunes, and this distorted a lot the result. It would be great to have a new bunch of articles to label, I think. -Theklan · · 19:44, 30 urtarrila 2019 (UTC)

Theklan OK it'll take some work to generate a new sample. I think what we'll want to do is try to build up a stratified sample of articles based on the predictions that are coming out of the current version of the model. I'm not quite sure how we'll do that yet, so I'm going to file it as a task. https://phabricator.wikimedia.org/T215351 I recorded some thoughts there. --EpochFail (eztabaida) 21:21, 5 otsaila 2019 (UTC)