Talk:WikiLoop

From Meta, a Wikimedia project coordination wiki
Jump to navigation Jump to search

WikiLoop Battlefield[edit]

Hello, I have serious concerns about WikiLoop Battlefield because it cannot deal with cases where a single user has made more than one consecutive vandalism edit. I believe it should be disabled until this is fixed, because at the moment it is doing more harm than good by only partially reverting vandalism in these cases. The` example I found this on was Federation of Australia on the English Wikipedia; see my message on Satdeep Gill's English Wikipedia talk page. Also, its edit summary on the English Wikipedia should really link to the English Wikipedia page on the software, not this Meta page. I hope this is a good place to provide feedback. Graham87 (talk) 04:27, 4 August 2019 (UTC)

I've reported the former issue on GitHub. Graham87 (talk) 07:31, 4 August 2019 (UTC)
Hi @Grahamwp(@Graham87:) thanks for your report of issue. Per your request, I currently disable revert. Let's have follow up discussion on Github issue about how we can resolve this problem. You should see the application behavior already changed. Xinbenlv (talk) 21:06, 4 August 2019 (UTC)

"Battlefield"?[edit]

The name may be unfitting for long-term use. Was "Antivandalism" or "Patrol" not fancy enough? ToBeFree (talk) 23:20, 30 December 2019 (UTC)

@ToBeFree: Hi thank you for reaching out, you are certainly not alone. We have been poked and pinged by Wikipedian friends and realized Battlefield is not a suitable name. We are currently asking for a brainstorm of a new name for this project, join the discussion here
Xinbenlv (talk) 23:47, 30 December 2019 (UTC)
Ah, thanks! ToBeFree (talk) 20:13, 31 December 2019 (UTC)

Can we now get rid of the voting banner?![edit]

Wikiloop does still show up the voting reminder banner.... --CommanderWaterford (talk) 09:29, 18 July 2020 (UTC)

will do Xinbenlv (talk) 12:54, 18 July 2020 (UTC)
done, thank you @CommanderWaterford:. Xinbenlv (talk) 19:27, 20 July 2020 (UTC)

Recent changes to interactions pipeline, etc.[edit]

@Xinbenlv: thank you for your good and hard work on DoubleCheck. As I mentioned on the signup sheet, I am fascinated by such systems, and have been working on them for some time. I have many questions, in five broad groups:

1. Could you please describe the data used in the recentchanges and interactions API Typescript for presentation on the JudgementPanel? Do you use anything other than the ORES scores? How about things like user account age or edit count, template tags on the article, article category membership, or recent article talk page history features? What are all the non-default RC query parameters?

Sure! describe the data used in the recentchanges and interactions API Typescript for presentation on the JudgementPanel: we mostly use the revision id, page article. We do use other vandalism detection scoring systems such as STiki score and ClueBotNG which is provided by STiki server, but it's currently down. We also receive WikiTrust system flagged revisions for review. These scores and suggested revisions-to-review are all only used as signals the reviewers to decide whether the revision is worth reviewing, these scores themselves doesn't affect Human judgements (except for potential psychological influence if any, but we try to design our new UI to minimize such influence).
Regarding How about things like user account age or edit count, template tags on the article, article category membership, or recent article talk page history features? What are all the non-default RC query parameters?, no we don't directly use such information as of now. But we do start to use article category for suggesting revisions-to-review. EllenCT (talk) 22:05, 12 August 2020 (UTC)
Could you do us a favor, help copy edit and put these answers to the tool's Wikipage if you think it's worth reading for other WikiLoop DoubleCheck users? Xinbenlv (talk) 23:50, 12 August 2020 (UTC)
You're welcome to if you want -- do you have some kind of a cross-project transclusion, because I see the sign-up sheet mirrored between enwiki and meta? EllenCT (talk) 02:41, 13 August 2020 (UTC)
Not sure whether it was done by cross-project transclusion, can we actually ensure these are transcluded? Maria helped did it. Xinbenlv (talk) 03:34, 14 August 2020 (UTC)

2. What are all the kinds of statistical processing, weighting, ranking, filtering, and sorting that occurs between RC and JudgementPanel presentation at present? What are your plans for enhancements? Or are you hoping to let it run as-is for a while so you can monitor it before deciding on future plans? EllenCT (talk) 22:05, 12 August 2020 (UTC)

Feature wise, WikiLoop DoubleCheck as a tool try to avoid creating its own ML scoring or statistically processing systems, but rather use 3rd party options in the ecosystem. We do plan to improve filtering and sorting. Xinbenlv (talk) 23:50, 12 August 2020 (UTC)
@Xinbenlv: I recommend that you filter out any edits involving wikidata numeric codes. EllenCT (talk) 02:31, 14 August 2020 (UTC)
Doing it in https://github.com/google/wikiloop-doublecheck/issues/325. Xinbenlv (talk) 03:34, 14 August 2020 (UTC)

3. How do you feel about the idea of vector space representations such as Wikipedia2Vec[1] (third party perspective, demo) generally? Do you see any possibilities for integration with DoubleCheck? Are you willing to speculate about how vector space concept models might be able to support possible enhancements? I have several ideas about this, some of which likely have more merit than others, so I don't want to jump into describing any of them until I get an idea of your familiarity and comfort with this area. EllenCT (talk) 22:05, 12 August 2020 (UTC)

I like the idea, and we look forward to integrating other scoring systems developed by the community, the academia or the industry. Xinbenlv (talk) 23:50, 12 August 2020 (UTC)
Thanks, I have a specific proposal involving the Kleinberg et al (2016) paper mentioned below. I want to treat edits as that paper treats people, and organize their "groups" as different sides of various controversial issues, in order to establish, measure, and visualize the set of relative biases across a given concept vector. I will tell you more about this once I remember a project I read about last year that was starting to use Wikipedia2Vec for an unrelated but partially similar task.... EllenCT (talk) 14:37, 13 August 2020 (UTC)
Em.. sounds interesting. I love the idea. I have also tried to do it, but so far I think the infra to do such big data analysis is missing yet .e.g a much better way to. Xinbenlv (talk) 03:34, 14 August 2020 (UTC)
How about instead of trying to extract information from the diffs, using the before and after texts of the edit, maybe with a few words, sentences, or paragraphs on either side of the actual diffs, as the basis for a concept vector representing that diff?
I know there are different ways to aggregate the individual words' and concepts, and that it needs to be done ideally on a per-predicate basis, which requires some kind of parsing across phrases. I'm still looking closely as how this might treat completely different articles instead of just small diffs.
I understand LSTM and other RNN solutions have outperformed symbolic parsing with sufficient training data, and that sentiment prediction remains relatively terrible, in part because it is subjective and varies by reader situation, in part because there are always going to be idioms for negation of different levels of obscurity, and likely for other linguistic reasons. EllenCT (talk) 19:46, 15 August 2020 (UTC)

4. Is there a formal interlingua that you have decided you want to use? How about Wikidata entries or related databases? I know that your colleagues are working on fully multilingual interlingual representations, and frankly I'm a little skeptical of that work in favor of projects such as yours (but I am on record saying I expect my opinion will change in 5-10 years), so I hope there is no need to commit to any particular formalism early. EllenCT (talk) 22:05, 12 August 2020 (UTC)

We hope to cover wikidata.org, but it will take a bit more time to design features for them. We don't currently plan to expand to non-Wikimedia projects or databases. I am not super familiar with the concepts of interlingua in your question. Feel free to clarify your question if my answer doesn't make sense to you.Xinbenlv (talk) 23:50, 12 August 2020 (UTC)

5. Finally, and most important, pertaining to for example [2], would you please describe the relationship between [3] and [4]; in particular how the trade-offs described in the former influence the characteristics described in the latter? Thank you for your kind attention to these questions. EllenCT (talk) 22:05, 12 August 2020 (UTC)

Sorry I am not super familiar with these two academic topic. I might have worked on relevant topics but mostly from a completely different angles Xinbenlv (talk) 23:50, 12 August 2020 (UTC)
[5] is an example application pertinent to the task; except, as above, edits are treated as people. EllenCT (talk) 02:21, 14 August 2020 (UTC)
Sounds interesting, would be interested in learning more applications on edits.Xinbenlv (talk) 03:34, 14 August 2020 (UTC)
Imagine that instead of people let into a school, you are judging edits let in to the encyclopedias. EllenCT (talk) 06:02, 14 August 2020 (UTC)

WMF Labs version of DoubleCheck is down[edit]

FYI: https://wmf.doublecheck.wikiloop.org/ is currently offline. The normal version still loads fine. — pythoncoder  (talk | contribs) 01:51, 10 September 2020 (UTC)

Bug reports[edit]

Is there an alternate location to report issues? I don’t want to create a Github account just for the purpose. What I would have reported is only minor: "Opened URL to revert not working" (Revert now button changes but article doesn’t open) – am logged in with WM OAuth, iOS 12 Safari. Pelagic (talk) 11:13, 23 October 2020 (UTC)

Notification problem on enwiki?[edit]

Hi, as an English Wikipedia user, I received a notification, which in its entirety reads:

"I'm writing to let you know we have simplified the RfC on trust levels for the tool WikiLoop DoubleCheck. Please join and share your thoughts about this feature! We made this change after hearing users' comments on the first RfC being too complicated. I hope that you can participate this time around, giving your feedback on this new feature for WikiLoop DoubleCheck users.
Thanks and see you around online,
María Cruz
MediaWiki message delivery (talk) 20:05, 19 November 2020 (UTC)
If you would like to update your settings to change the wiki where you receive these messages, please do so here."

Note the link to en:Global message delivery/Targets/WLDC Users, which is a mainspace page at the English Wikipedia and is redlinked. Can this be fixed? Thanks. Eumat114 (talk) 01:29, 20 November 2020 (UTC)

Can we add a feature that allows the revision to be opened with just one key?[edit]

I use Wikiloop, Twinkle and RedWarn (and some others) to combat vandalism. The problem is, said anti-vandal tools only work if I'm looking at a diff of the vandal's edits. Thus, I propose using the up arrow key as a shortcut for opening the diff. This would make anti-vandalism far easier for me and various others. Thanks for your time. Opalzukor (talk) 21:07, 11 December 2020 (UTC)

Garbage[edit]

  • How is this "Identified as test/vandalism"? Fix your garbage vandalism detection algo please.
  • @JeffyBoeu: check your edits when using this tool. Tom.Reding (talk) 19:08, 26 March 2021 (UTC)
  • @Tom.Reding: sorry about that. I'm new to this. JeffBoeuu (talk) 2:15, 26 March 2021 (MDT)

Where can I find instructions or a guide to WikiLoop DoubleCheck?[edit]

I'm looking for a guide that answers questions like these:

  • What is "ORES Damaging" and how does one interpret the percentages?
  • Similarly, what is "ORES Bad faith" and how does one interpret the percentages?
  • If I go to the article and fix something identified by WikiLoop DoubleCheck, should I do anything extra to document those edits? For example, an editor identified vandalism, but he/she/they did not know what should replace the vandal's content so they substituted "Fix plz". My edits (diff) followed.

Thanks! Mark D Worthen PsyD (talk) [he/his/him] 05:50, 4 April 2021 (UTC)

Hi User:Markworthen, You can find information about ORES in its Mediawiki article https://m.mediawiki.org/wiki/ORES and in its FAQs subarticle. About the last question, it is recommended to warn the user/vandal of the revert. The "easy" way to do it is using a tool like Twinkle. In desktop mode appears a menu (TW) and while watching the user page, the option "Warn" must be selected then appears a list to select the warning level and the reason of the revert. It is all better explained in the links provided :-) Alexcalamaro (talk) 09:08, 4 April 2021 (UTC)
Awesome, thanks so much Alexcalamaro! Mark D Worthen PsyD (talk) [he/his/him] 02:53, 5 April 2021 (UTC)