Missing thread?
See the /Archive

Volunteer retention research edit

Been learning a lot over the years about the nature of volunteering in the wiki communities, and am digging in to help improve volunteer recruitment and retention methods. Would love to connect with you for some Q&A to better understand the research you've done in the past and where you see the biggest possibilities and challenges now. Email me to figure out when we could talk? Thanks! DrMel (talk) 23:02, 8 December 2016 (UTC)Reply

Do we have any data on the gender makeup of early editor loss, the people who never make it to 10 or 100 edits? Did the gender makeup of editor retention change in 2007-ish? Forgive me if I've overlooked this information somewhere. HLHJ (talk) 02:56, 8 April 2018 (UTC)Reply

Regretfully, I don't know the answer to this. I've been trying to get a proper funnel analysis of early editing (and pre-editing) dynamics in place for a while. It would be great to know where the drop-offs are around gender and other demographics. I have not been able to get the work prioritized. However, I know that LZia is looking into expanding our research around diversity of contributors next year. Maybe this is something that is on her plate. --Halfak (WMF) (talk) 14:57, 12 April 2018 (UTC)Reply
Forgot to ping HLHJ --Halfak (WMF) (talk) 14:57, 12 April 2018 (UTC)Reply
Thanks, Halfak (WMF) for looping me in.
Hi HLHJ, Hi DrMel. A few pointers to share here:
  • We're currently working on growing contributor diversity as part of the current annual plan (the program). We're done with Objective 1 and 2, and we are working on Objective 3 this quarter (April-June). If you want to get pings on the status of Objective 3, please follow Phabricator:T190776 and Phabricator:T190775. If you want more detailed information about each of these tasks, please check the research on eliciting new editors' interests as well as designing a framework for increasing the retention of women in Wikipedia's editor pool. The parent research for both of these efforts where you can find some literate review may also be interesting for you.
  • The above is current year's work and this year ends at the end of June 2018 for us. What we have proposed to do as part of next year's annual plan (pending FDC+Board review, recommendation, approval) is:
  • Finalize the model for editor-editor matching recommendation. Think of this as something that can be helpful for both newcomers and experienced editors: what if we had a system where you could get recommendations about what editors have similar interests to you? Or, what if we could learn about the interests of newcomers fast, without having to wait for them to interact with our systems for a long time?
  • Further tune and test the framework that we hope to help us keep a more diverse editor pool on Wikipedia (and possible other Wikimedia projects).
  • Create baseline statistics that can help us understand the map of diversity at many stages of the editing funnel, something Halfak mentions above.
The pointer to the annual plan is available for your review. Please check Output 2.2 (which is focused more on readers) and Output 3.1 and 3.2 (continuation of this year's research linked above), and Output 3.3 where we want to create a map of the different stages of the editing to understand the state of the diversity better. Note that this is beyond gender diversity and I'd like for us to at least try to include ethnicity in it as well.
I hope this overview helps. And please keep in touch as you seem to be interested in this space. :) --LZia (WMF) (talk) 19:34, 12 April 2018 (UTC)Reply
Thanks to Halfak and LZia (WMF). I'm sorry, I managed to miss this notification. There is an interesting intersection here. If, as Research:Voice and exit in a voluntary work environment says, we are short on women because women have less self-confidence and are more likely to shy away from conflict, then new editors leaving because they face hostility and are made to feel useless are more likely to be female. The research suggests that rejection and futility discourage editors. So the gender bias in editor retention should correlate with the editor retention rate, assuming the theory is at least partly right. A parallel hypothesis for non-dominant cultures would predict a parallel correlation.
I really appreciate the efforts being made on this. I'm afraid I'm going to be critical here, please take it as attempted support rather than hostility for your efforts. From personal experience, I've never felt any need of help finding things to edit; I could keep a dozen of me busy. Nor did I want to interact with people; there are better places to do that. I wanted to make useful edits. If I'd felt more insecure, I might have wanted explicit reassurance that my edits were useful, like, say, thanks notifications. Making useful edits was much easier before automated tools made it easier to revert edits than fix them. The research seems to me to suggest that we need ways to increase fixing and decrease reverting of good-faith edits by new users. While focussing on modifying the behaviour of new editors is easier politically, I suspect it will not be as effective as modifying the behaviour of established editors. HLHJ (talk) 02:40, 22 October 2018 (UTC)Reply

Possibly investigate data regarding Single Edit Tab and Visual Editor edit

Some time ago Single Edit Tab was deployed on EnWiki, with a default placing all new users into VE first. The default was latter reversed, starting all new users in the Wikitext editor. This was not a controlled experiment, but it might be worth checking whether any noticeable signal jumps out at the point of initial deployment or at the point that the default was reversed. I'd love to see a simple plot of the data over time, with critical points marked, just to eyeball whether there was some visible jump. We could look at the percentage of editors who successfully make a first edit, editor retention, or anything else you think might be relevant.

Does this sound worth doing? Alsee (talk) 14:31, 5 February 2017 (UTC)Reply

Hey Alsee. I'm not too familiar with the history of these deployments since I haven't been actively involved in the development of VE for a while. Do you know the dates of deployment? What kind of stats would you want to look at during these time periods? I wonder if Neil P. Quinn-WMF has already done some exploration. --Halfak (WMF) (talk) 15:33, 6 February 2017 (UTC)Reply
@Alsee: No, I haven't explored that. It's an interesting question but I'm not sure looking at the graphs would be all that illuminating since there's a lot of baseline variation in that data. You're welcome to file a task requesting that analysis by following the instructions at mw:Editing/Analysis, but the reality is that I have a very long backlog and I'm not likely to be able to do it any time soon.—Neil P. Quinn-WMF (talk) 18:17, 6 February 2017 (UTC)Reply
It's possible that we could so something basic, but useful, with a few Quarry queries. Alsee, are you familiar with Quarry? I'd be happy to give you a hand with picking up some of the basics. --Halfak (WMF) (talk) 22:57, 6 February 2017 (UTC)Reply
@Neil P. Quinn-WMF: and Halfak. Thanx.
  • Here's the Phab link[1] for the Single Edit Tab deployment. Successful gerrit merge[2] Apr 12 2016, 11:02 AM. It was definitely live on EnWiki prior to Apr 12 2016, 4:30 PM. A 5 hour window at most.
  • Here's the Phab link[3] for changing the VE-default to a Wikitext-default. Successful gerrit merge[4] on May 19 2016, 7:22 PM. I'm not sure if that's when it went live on EnWiki, or if it was soon after that.
I know any signal will probably be lost in the noise, but if there is a signal big enough to spot then it's a signal that's worth knowing. The bonus here is that this is a two-for-one deal. One search can span the two deployment dates, and two significantly different potential signals. I'm flexible on what to look at, but my general thought was some of the same basic data that were checked in the May 2015 Visual Editor experiment. For new accounts: what percentage made at least one edit (and maybe at least 5 edits?), one week editor survival (I think 3 month survival was checked?), total edits. Whatever is easy to pull out of the database, whatever you have time for, whatever you think might turn up meaningful data.
I would love to pick up Quarry skills, and as a programmer I'm sure I could. The examples I've seen are semi-readable even with zero knowledge. I don't think I can dive into it now though. Do you have a good link on it that I can bookmark? Alsee (talk) 00:53, 7 February 2017 (UTC)Reply
@Alsee: Hey Alsee. Sorry for the short response, but since you are a programmer, yes, I'm sure you'd be able to deal with Quarry quite easily! The documentation is located at Research:Quarry and provides a decent introduction to the basic Quarry-specific bits. Other than that, you just have to know (1) SQL, which you can learn in about a thousand places online if you're not familiar with it—I've used the Tech on the Net tutorial before and it's pretty good and (2) the way the MediaWiki database is laid out, which is documented at mw:Manual:Database layout—as it says there, the most important tables for you are page, revision, and user tables.Neil P. Quinn-WMF (talk) 02:12, 23 February 2017 (UTC)Reply

Your feedback matters: Final reminder to take the global Wikimedia survey edit

(Sorry to write in Engilsh)

A barnstar for you! edit

  The Original Barnstar
Thank you for your interesting lightning talk at the Wikimedia Conference 2017. Ijon (talk) 08:43, 4 April 2017 (UTC)Reply

Update regarding Wikipedia and Wikimedia Commons tutorial videos edit


I regret to inform you that the series of motivational and educational videos project, which had been planned introduce Wikipedia and some of its sister projects to new contributors, is being discontinued.

There are multiple factors that have led to this decision. The initial budget and time estimates were far too small for a project of this scale and complexity. Also, my simultaneous involvement in Cascadia Wikimedians User Group was problematic due to the shortage of human resources for the user group, which resulted in my spending far more time trying to help the user group than I had planned, so my time and attention were diverted from this video project to assisting the user group.

You can find more information in the final report for the grant.

I regret that this project did not fulfill the hopes that many of us had for it, and I hope that in the future someone with the necessary resources will choose to resume work on it or a similar project. If you are interested in working on this or a similar project then please contact the WMF grants team.

On a personal note, I am retiring from the Wikimedia community. Perhaps I will return someday.


--Pine 23:20, 30 August 2017 (UTC) Reply

Series director and screenwriter

Join ORES project edit


I've just discovered New Developers page now. And I want to know if it is possible to join the project.BamLifa (talk)

Hi BamLifa! Yes. You're very welcome. The best way to work with us to to find us in IRC. See #wikimedia-aiconnect. I'm 'halfak'. 'Amir1' and 'awight' should also be able to help you get started if I'm not around. --Halfak (WMF) (talk) 18:42, 1 September 2017 (UTC)Reply

Hey Aaron edit

Hello, Aaron, i thought i let you know that your userpage is a bit outdated, as you still are mentioning the volunteer project, Revision scoring as a service, which is now obviously Scoring Platform Team. Anyway, have a great week! Zppix (talk) 19:26, 7 November 2017 (UTC)Reply

  Done! Thanks. --Halfak (WMF) (talk) 11:18, 13 November 2017 (UTC)Reply

Oauth Consumer proposal edit

Hello Halfak! Sometime ago, you approved my oauth application Chlambase.org with your EpochFail account. I have created another model organism database for a new pathogen, Myxococcus xanthus and I could use your help again to get my oauth application approved! It should be identical to Chlambase.org. Here is the proposal. Thank you for any help! Derek (talk) 18:42, 30 August 2018 (UTC) @EpochFail:Reply

Just got back from Vacation.   Done BTW, I'm still active as EpochFail and my work as an OAuth admin is all volunteer so I do it from that account.  :) --Halfak (WMF) (talk) 14:35, 4 September 2018 (UTC)Reply

Mail edit

--Rosiestep (talk) 14:52, 20 November 2018 (UTC)Reply

Your Thoughts edit

I noticed that you are the Foundation contact for Research:How much do Wikipedians in the US value editing Wikipedia? There has been resistance to the talk page requests sent by the researcher at their English WP talk page. Would you or the foundation have anything to add to the discussion there? Thanks! -- Dolotta (talk) 16:32, 19 December 2018 (UTC)Reply

Hi Dolotta! I've responded from my volunteer account, User:EpochFail, since I'm not working on this in my official role as a staff member. Thanks for your ping. --Halfak (WMF) (talk) 20:58, 19 December 2018 (UTC)Reply
Thanks for letting me know. Happy editing! -- Dolotta (talk) 22:43, 19 December 2018 (UTC)Reply

Question regarding technical research for Wikivoyage edit

Hello. Apologies if you are not the right person to whom I should pose this question.
I am an administrator at the English Wikivoyage. There, we are considering a proposal to supplement redlinked articles with an automatic link to the corresponding Wikipedia article. However, there was some concern that adding such a link would disincentivize creations of these articles locally. We are a relatively small community, and creations of articles for travel destinations by new or anonymous users are a regular occurrence, and quite valued. We were hoping it would be possible to gather some kind of data on this. Beyond making the change for a trial period and comparing the number of page creations before and during, none of us have any ideas. Of course, not all page creations come from redlinks, and so a more sophisticated test would be ideal. Is it possible to do this within the Research framework? If so, what would we need to do?
Any help or resources you can give us would be immensely appreciated. Thanks, ARR8 (talk) 01:12, 31 March 2019 (UTC)Reply

Hi ARR8! I could possibly help you with this. Generally, I would recommend running a trial for whole week long periods to address the periodic nature of week days and weekends. Just how many weeks you'd need to run the trial for in order to see if there is an effect is a more complicated question that I'd need to do some analysis to explore. Essentially we want to know how long the trial would need to run in order for us to get enough data to find any substantial change to be statistically significant. Do you have a sense for how much of a change would be meaningful? E.g. if article creation rates dropped off by 5%, would that be too much? Or would it need to be something more like 20% in order to matter? --Halfak (WMF) (talk) 21:10, 31 March 2019 (UTC)Reply
Thank you! My guess is that ~8-12% would be a good thershold (corresponding to around half a standard deviation by some very cursory statistical measures), but some of our editors agree that 5% would be noticeable. ARR8 (talk) 03:20, 1 April 2019 (UTC)Reply
ARR8 Can you show me how you arrived at that standard deviation? I could probably use that to do a statistical power analysis. --Halfak (WMF) (talk) 13:44, 1 April 2019 (UTC)Reply
Sure. It's not a very meaningful statistic, but I used the auto-generated number of content page creations here (which may be your own creation?) going back two years. This would include all articles by every editor, though, not just destinations by new users, and maybe also includes non-mainspace page creations. ARR8 (talk) 15:33, 1 April 2019 (UTC)Reply
ARR8 I see. This is useful actually we can make some good estimates from it. One thing that is strange though is that it doesn't look like there are *any* anon page creations. Could that be a bug with the data in this tool? (The tool is not my work but I helped a bit). --Halfak (WMF) (talk) 18:35, 1 April 2019 (UTC)Reply
Definitely a bug. One of our editors came up with another set of numbers using a different tool, listing only destination articles. Maybe this is more helpful? ARR8 (talk) 22:41, 1 April 2019 (UTC)Reply
I just tried my own query. https://quarry.wmflabs.org/query/34844 gets the daily article creations by anons for the last 30 days. I get a mean of 40 articles and a standard deviation of 22.7. If we ran the trial for 2 months, we could find statistical significance for changes bigger than + or - 8 articles per day (32 - 48 avg). We'd have to run the trial for a whole extra month to notice differences smaller than thank + or - 7.
That said, we could see if there is a huge change (e.g. going from averaging 40 articles per day from anons to 20 articles per day from anons), in 1-2 weeks.
Would that fit into the timescale y'all are thinking about for the experiment? --Halfak (WMF) (talk) 22:57, 1 April 2019 (UTC)Reply
Sorry for the delay. Those times work fine, and I think the community would choose to run it for the extra month. Thanks again. What are our next steps? ARR8 (talk) 01:52, 5 April 2019 (UTC)Reply


ARR8 Oh! One other question. I'm concerned about seasonal fluctuations. We often see changes in editing behavior due to kids going on summer vacation/going back to school. Or due to the winter holiday season. Judging from the graph of page creations, it looks like there aren't any major seasonal fluctuations that we should be worried about. The graph looks mostly flat from 2014 forward (except for a minor dip in early 2016 that seems to not repeat).
Otherwise, I think the next step is to schedule the change you want to make. From there we can project out when we'll want to do analysis to see if you want to reverse the change. I'm thinking that we'll want to have a look two weeks in, one month in, and then two months in assuming all is going well by that point. It shouldn't be too much trouble for me to run some basic analysis for you at those points. Just confirm the dates with me before you kick it off and I'll make sure I'm available. --Halfak (WMF) (talk) 13:31, 5 April 2019 (UTC)Reply
I agree that seasonal variations don't seem to be a major factor. One source of concern brought up in the community, though, was that measuring anonymous-user page creations may not be a perfect proxy for redlinks-turned-into-articles, and it was requested that this be directly measured. Is this possible, or necessary?
In either case, we're ready to switch the template at any time. We have consensus to run the test. Let's say tomorrow, assuming nothing further can/should be done; how's that sound? ARR8 (talk) 03:49, 7 April 2019 (UTC)Reply
ARR8: I'm not sure how I'd measure red links turned into articles. It turns out that's hard to get from the database. Generally, I'm thinking that if most anon-created articles come from redlinks, then any effect will be visible in the total article creations. If a small proportion of anon article creations come from redlinks, then (and this is a bit hazardous to say) maybe it doesn't matter if the automatic link causes a drop in such article creations. Either way, if it proves to be essentially, I have some ideas about how we might get at it after-the-fact. Otherwise, I think now is a fine time to kick off the change. Please note the time when you make the change here so I can reference it later :) --Halfak (WMF) (talk) 23:23, 10 April 2019 (UTC)Reply
Sure. I made the change yesterday: diff and timestamp here. Thanks again. ARR8 (talk) 21:28, 16 April 2019 (UTC)Reply

Hi, any update on this? ARR8 (talk) 23:30, 16 May 2019 (UTC)Reply

ARR8, I can try to take a look at this next week. I'll post an update here when I get some analysis done. :) --Halfak (WMF) (talk) 07:50, 17 May 2019 (UTC)Reply
I started work here: Research:Measuring the effect of cross-linking missing articles in English Wikivoyage --Halfak (WMF) (talk) 14:53, 24 May 2019 (UTC)Reply
I have a major update. Basically, the effect is currently too small to be differentiated statistically. But it looks like, on average, we're seeing 1.3 fewer anonymous article creations per week since the change was made. If we let the trial continue for a few more weeks, we might be able to find a statistical significance, or the observed effect may wash out. No reason to conclude any meaningful difference yet. See Research_talk:Measuring_the_effect_of_cross-linking_missing_articles_in_English_Wikivoyage/Work_log/2019-06-03 --Halfak (WMF) (talk) 22:00, 3 June 2019 (UTC)Reply
Forgot to ping ARR8. :) --Halfak (WMF) (talk) 13:17, 4 June 2019 (UTC)Reply
ARR8, I just did an update with more data and it still looks like the difference is not significant (p = 0.31). It might be safe to conclude by this point that the change did not substantially affect anonymous page creations. --Halfak (WMF) (talk) 16:41, 17 July 2019 (UTC)Reply
I just fixed a minor bug where I was measuring a partial week and the p value got even less significant (p = 0.46) --Halfak (WMF) (talk) 16:43, 17 July 2019 (UTC)Reply
I also realized that I should add for clarity that the wiki seems to be down about 0.24 new anon-created articles per week on average. That's such a small change that it's almost certainly due to chance. --Halfak (WMF) (talk) 19:32, 17 July 2019 (UTC)Reply
Thank you very much! ARR8 (talk) 02:12, 21 July 2019 (UTC)Reply

Editor retention research idea edit

I know you have an interest and expertise in research related to editor retention. I'd like to toss out an idea, if you're interested.

For several years there have been serious concerns about abusive admins at AzWiki. For example RFC: Do something about azwiki, although I believe the closure-statement on that RFC seriously understates the situation. There are many issues, which seems to include a pattern of spamming unexplained reverts (often as rollbacks) or abusive blocks to defeat or drive away anyone they disagree with. The admins collectively have a pattern of ultra-nationalism and genocide denial. One was quoted in a news item as advocating Wikipedia for "Information warfare". This admin insta-blocked a visiting EnWiki admin who tried to move the article "Fictional Armenian Genocide" to the the title "Armenian Genocide". This admin has been removed for other reasons, but the problem is systemic.

It's hard to get a handle on just how widespread the blocking and unexplained-reverts issues are, but I think it would help to get data. I would expect unexplained reverts, especially chains unexplained reverts, to be particularly destructive on new editors. I saw another experienced editor from EnWiki try to contribute there. They got hit with a chain of at least 6 unexplained reverts from an admin, and then blocked. There's no way we could expect to retain any new user who faced that sort of abuse. I think it woud be valuable to have data on a clearly unhealthy wiki. Any kind of community health stats may be useful. Maybe we can catch other wikis that are in trouble, spot wikis which may be heading in a bad direction, or maybe we can get data or ideas to help to steer "normal" communities in a more positive direction. And of course the data might help us improve things at AzWiki.

I know this wouldn't be a trivial project. If it doesn't interest you, no problem. The global community is floundering for useful information and solutions, and I thought this might help. Alsee (talk) 18:03, 20 July 2019 (UTC)Reply

I belatedly realized that Talk:Community health initiative appears to be an excellent fit for this. Should I post it there insead? Alsee (talk) 18:28, 20 July 2019 (UTC)Reply

Thanks for the information and request, Alsee. This is interesting to me. I don't think we have a good way to get long term community health metrics per wiki from any sort of dashboard, but I sure do want that. I think that, at the very least, we could look at the community growth/retention dynamics like we did in the R:The Rise and Decline to see what is happening there. It shouldn't take too much work to look at revert rates and retention. More specifically, I'm interested in these reverts without explanation. Can you tell me more about what those look like and maybe give me an example or two? I'm also curious if you think there was any sort of sudden shift in the way newcomers are treated or if it was more of a progressive change over time. --Halfak (WMF) (talk) 20:29, 24 July 2019 (UTC)Reply
Mostly I just know that there have been a lot of serious complaints for at least a few years of systematic bad behavior by AzWiki admins, and adding my personal recent observations of an apparently significant patterns of revert behavior. I doubt there was any sudden change. The admins appear to use the rollback button a lot, which does not allow any option to alter the default edit summary. When it's not a rollback, it would generally be a revert with the default edit summary unchanged:
  • Here's an example of three rollbacks in a row by an admin on one article,[5][6][7] interweaved with three rollbacks in a row by the same admin on another article.[8][9][10] That's quite blatant edit warring - then the admin hit the other editor with a block. No discussion, no warning, no nothing. Here's a pair of rollbacks in a row by another admin.[11][12] The rollbacks have Tag:Rollback and the edit summary format is: "USERNAME-LINK tərəfindən edilmiş redaktələr geri qaytarılaraq USERNAME-LINK tərəfindən yaradılan sonuncu versiya bərpa olundu."
  • Here's an example of a revert with the default edit summary. It has Tag:Undo and the default edit summary format is: "USERNAME_LINK(TALK_LINK) tərəfindən edilmiş REVISION# dəyişikliyi geri qaytarıldı."
I am merely speculating, but I would guess that unexplained reverts are propbably common among non-admins too. The admins probably either serve as a role model for community behavior, or perhapse their behavior may reflect common community behavior. (Or more likely both, in a circular fashion.) Alsee (talk) 13:00, 27 July 2019 (UTC)Reply
Alsee Thanks for the details. The examples are really helpful. I'm wondering if there would be a useful metric we could generate around unexplained reverts. This might be a useful metric for MMiller and the mw:Growth team. Essentially, good quality control practices (e.g., explaining revert reasons) should be part of improvements to newcomer socialization practices. We should probably track these kind of metrics both to target interventions and track progress. I think that Analytics' MediaWiki edit history tables will make generating these metrics pretty straightforward because they already track reverts. --Halfak (WMF) (talk) 15:26, 6 August 2019 (UTC)Reply
I'm wondering if there would be a useful metric we could generate around unexplained reverts. - I know that you know the available data, and what you can do with it, a far better better than I do. But for what it's worth here is what I picture:
First, I'd essentially define "reverts" as "reverts + rollbacks". The technical distinction between reverts and rollbacks doesn't much matter to the person who had their edit undone.
I'd essentially define "unexplained reverts" as "default-edit-summary-reverts + rollbacks".
Then I'd look at the rate for each type, where rate = edits-of-that-type divided by all-edits-to-that-wiki, in a given time frame.
This one is largely motivated by AzWiki, but I'd also be interested the rate-of-each-type-made-by-admins divided by all-admin-edits, in a given time frame.
What percentage of reverts are unexplained reverts?
Then I'd want to look across many wikis (or all wikis) to see what was "normal" for these figures and what outliers there are. And of course compare those figures with retention rates for new users.
The impressive fancy magic would be if you look at the fate of individual editors who were hit by these kinds of edits. My expectation is that unexplained reverts, especially multiple unexplained reverts, would hit new users really hard. They get no useful information or engagement. A revert with a cryptic-acronym link to a policy page at least gives them some chance to understand why their edit was reverted. It gives them a chance to start learning that we have policies and how they work. It gives at least some sense of control to try and avoid future reverts. They also begin to learn that they can cite those policies to gain power and effectiveness as editors. That's a big part of becoming a successful editor. Oh! The Wikipedia article Learned helplessness is probably extremely on-point for unexplained reverts:
Learned helplessness is behaviour exhibited by a subject after enduring repeated aversive stimuli beyond their control. It is characterized by the subject's acceptance of their powerlessness: discontinuing attempts to escape or avoid the aversive stimulus, even when such alternatives are unambiguously presented. Upon exhibiting such behavior, the subject is said to have acquired learned helplessness.
In humans, learned helplessness is related to the concept of self-efficacy, the individual's belief in their innate ability to achieve goals. Learned helplessness theory is the view that clinical depression and related mental illnesses may result from such real or perceived absence of control over the outcome of a situation.
In other words, new users learn to give up. Alsee (talk) 09:14, 7 August 2019 (UTC)Reply

Approx deletion rate for established articles edit

Hi, I'm trying to estimate the approx deletion rate for established Wikipedia pages (initially posted here). Are there stats anywhere for the number of established articles deleted each day (i.e. not via w:WP:AfC/w:WP:NPP). There are some great stats at Research:Wikipedia_article_creation about articles deleted soon after their creation, but I've not managed to find anything on established pages (more than a few months old). Any rough useful, detailed stats a bonus! T.Shafee(Evo﹠Evo)talk 05:57, 29 August 2019 (UTC)Reply

Evolution_and_evolvability Sorry for the delay! I was on vacation. I don't know of any easily accessible stats on the deletion rate for established pages. But it seems like we could get them somehow. E.g., we could define an established articles as a mainspace page that is not a redirect or disambig that survives for at least 1 month (given my analysis of deletion after creation) in mainspace. Given my past work, I expect the survival rate of this group to be quite high, but that the deletion reasons will be very interesting. I don't have the time to invest in something like this right now, but I'd be happy to advise if we can get someone who wants to run a lot of SQL to work it out. --Halfak (WMF) (talk) 14:29, 4 September 2019 (UTC)Reply

ORES in Turkish edit

Hello Halfak. I am Evrifaessa from trwiki. I am interested in training ORES' AI in Turkish with the community of Turkish Wikipedians.

I see that the progress is just at 10% right now, I'm planning to form a contest-like thing for trwiki users so we can come together and start training the AI to make it work in our language. I made a demo to see how accurate it is right now, but I saw that it's not as accurate as I thought it would be. And also, the dataset that's in use right now in here belongs to 2016. Can we update it to 2020 somehow?

I'll be waiting for your answers, please tell me the best ways to start using ORES in Turkish. Have a nice day :)--evrifaessatalk 09:02, 2 July 2020 (UTC)Reply

Lemme ping other users of Wikimedia Scoring Platform team. @ACraze (WMF):, @KBazira (WMF): and @Haksoat:. Our community is waiting for your response.--evrifaessatalk 14:09, 3 July 2020 (UTC)Reply
Hello Evrifaessa! Thanks for the ping. I've created Phab:T257359 to track the work. We'll ping back here when the new labeling campaign is ready. --Halfak (WMF) (talk) 19:02, 7 July 2020 (UTC)Reply
Hey Halfak, hello again. Will we be able to limit the usage rights of the labeling gadget only to patrollers in trwiki? We'd definitely not want some vandals to intentionally false-label revisions. That would decrease the accuracy of ORES for sure. Have a nice day.--evrifaessa ❯❯❯ mesaj 18:40, 25 July 2020 (UTC)Reply