Wikipedia talk:Community health initiative on English Wikipedia/Edit filter

This page is currently inactive and is retained for historical reference.
Either the page is no longer relevant or consensus on its purpose has become unclear. To revive discussion, seek broader input via a forum such as the village pump.

Exploring how the Edit filter can be used to combat harassment[edit]

The Edit filter (also known as AbuseFilter) is a feature that evaluates every submitted edit, along with other log actions, and checks them against community-defined rules. If a filter is triggered the edit may be rejected, tagged, logged, trigger a warning message, and/or revoke the user’s autoconfirmed status.

Currently there are 166 active filters on English Wikipedia. One example would be filter #80, “Link spamming” which identifies non-autoconfirmed users who have added external links to three or more mainspace pages within a 20 minute period. When triggered, it displays this warning to the user but allows them to save their changes. It also tags the edit with ‘possible link spam’ for future review. It’s triggered a dozen times every day and it appears that most offending users are ultimately blocked for spam.

AbuseFilter is a powerful tool at handling content issues and we believe it can be extended to handle more user conduct issues. The Anti-Harassment Tools software development team is looking into three major areas:

1. Improving its performance so more filters can run per edit

We want to make the AbuseFilter extension faster so more filters can be enabled without having to disable any other useful filters. We’re currently investigating the current performance in T161059. Once we better understand how it is currently performing we’ll create a plan to make it faster.

2. Evaluating the design and effectiveness of the warning messages

There is a filter — #50, “Shouting” — which warns when an unconfirmed user makes an edit to mainspace articles consisting solely of capital letters. (You can view the log if you’re curious about what types of edits successfully trip this filter.) When the edit is tripped, it displays a warning message to the user above the edit window:

These messages help dissuade users from making harmful edits. Sometimes requiring a user to take a brief pause is all it takes to avoid an uncivil incident.

We think the warning function is incredibly important but are curious if the presentation could be more effective. We’d like to work with any interested users to design a few variations so we can determine which placement (above the edit area, below, as a pop-up, etc.) visuals (icons, colors, font weights, etc.) and text most effectively conveys the intended message for each warning. Let us know if you have any ideas or if you’d like to participate!

3. Adding new functionality so more intricate filters can be crafted.

We’ve already received dozens of suggestions for functionality to add to AbuseFilter, but we need your help to winnow this list so we can effectively build filters that help combat harassment.

The first filter I propose would warn users when they publish blatantly aggressive messages on talk pages. The user would still be allowed to publish their desired message but the warning message would give them a second chance to contemplate that their uncivil words may have consequences. Many online discussion websites have this functionality, to positive effect. The simple version would be built to detect words from a predefined list, but if we integrated with ORES machine learning we could automatically detect bad faith talk page edits. (And as a bonus side effect, ORES could also be applied to content edit filters.)

Another filter I propose would be to log, warn, or prevent 3RR violations. This filter has been proposed twice before ([1], [2]) but rejected due to lack of discussion and because AbuseFilter cannot detect reverts. The Anti-Harassment Tools team would build this functionality as we believe this filter would provide immense usefulness in preventing small-scale harassment incidents from boiling over.

There are countless other filters that could be created. If you wanted to create a filter that logged, warned, or prevented harassing comments, what would it be? And what functionality would you add to AbuseFilter?

- The Anti-Harassment Tools team (posted by SPoore (WMF) (talk) , Community Advocate, Community health initiative (talk) 16:20, 22 June 2017 (UTC) )[reply]

Discuss if AbuseFilter has the potential to alleviate harassing behavior[edit]

Do you think AbuseFilter has the potential to alleviate harassing behavior?

...

Discuss design changes to the warning messages[edit]

What design changes to the warning messages should we explore?

Please stay away from pop-ups/pop-overs unless it can be controlled per-filter; these can be very disruptive to the editing process if there is a false positive. — xaosflux ^Talk 01:59, 22 June 2017 (UTC)[reply]

@Xaosflux: I agree, and I like your suggestion of allowing EFMs to select which type of warning. We want to perform some usability tests on a few designs so we can make an informed decision. If there is no noticeable difference then we'll keep it simple. — TBolliger (WMF) (talk) 17:28, 22 June 2017 (UTC)[reply]

Discuss existing filters which are effective[edit]

Which existing user conduct filters are effective?

...

Discuss potential new filters[edit]

If you wanted to create a filter that logged, warned, or prevented harassment, what would it be?

WP:HOUNDING filter[edit]

Editing more than 10 or 20 pages immediately after another user edits them within maybe 1 day. Count edits made within a day, not the length of time since the 1st editor made the edit. Such editing patterns are a strong indication of either hounding or some serious cleanup. The filter should trip a warning that will give the editor pause to consider their actions, but should allow the editor to override a warning. Adding the edits to a log would flag for Admin attention either against the vandal or the hounder.

At the 10th follow on edit a user talk message is generated that says "AbuseFilter has detected you have edited 10 pages immediately after User:X. Following a user from page to page can be considered WP:HOUNDING unless you are reverting obvious vandalism. Exercise caution before proceeding. " Legacypac (talk) 23:57, 22 June 2017 (UTC)[reply]

@Legacypac: A filter for wikihounding? I think that's a pretty strong idea to pursue. I agree that a warning would be the best response — in many cases users might not be aware of the policy or be caught up in the moment so a gentle reminder would give them a pause to make a conscious decision about their next actions. I'm curious what other folks think about this type of filter. — TBolliger (WMF) (talk) 17:41, 23 June 2017 (UTC)[reply]

This is an intriguing idea, but let me identify one potential false positive. I tried to start every day by knocking off 10 items at Copypatrol. In most cases, if I find a copyright violation I will rollback the edit. If I find 10, I might trigger this filter. Empirically, some of the edits I review are okay I marked them as acceptable and do not make an edit so my typical day probably includes only six or seven but on some days I do 15 or more and probably have 10 edits within an hour or so all of which are immediately after some other user edits. While my guess is that I will rarely trip the filter, my guess is that Diannaa who does about 10 times as many as I do, will be in constant violation. It may be as simple as adding a white list — for this particular type of edit there are realistically only two of us who are likely to trip it.

Another possible false positive with an easier solution is edits to the category Category:Non-free_files_with_orphaned_versions_more_than_7_days_old. It is quite common I will do a few hundred in a row. I believe in 100% of the cases the immediately prior editor is a bot, so it may be sufficient to design the edit filter that edits following a bot don't count.--S Philbrick (Talk) 21:26, 24 June 2017 (UTC)[reply]

Good points. Another false positive is any Admin working a CSD category. One day one admin deleted 173 pages I CSD tagged. I'd be happy to exclude all Admin usergroup from triggering the filter. One would hope Admins are not hounding for no reason. Legacypac (talk) 00:59, 25 June 2017 (UTC)[reply]

In most edit filters we exclude large known groups (in some cases all confirmed users, or users with over 500 edits, bot accounts - we can easily also exclude admins, page patrollers, or any other usergroup). — xaosflux ^Talk 01:28, 25 June 2017 (UTC)[reply]

Bots should be excluded too. Presumably bots don't wikihound :) All humans, except Admins because they are likely to trip the filter too often in the normal course. Legacypac (talk) 01:44, 25 June 2017 (UTC)[reply]

Great ideas. I agree that we'll want to set some requirements on the users involved. And since a Wikihounding filter would probably work best with a warning, we could also think about adding functionality to throttle the warnings (e.g. one show a warning one time to a user per day/week/year/lifetime.) This filter might work best as an one-time educational system. — TBolliger (WMF) (talk) 22:08, 26 June 2017 (UTC)[reply]

Trip filter at 10 subsiquent edit - get a warning. Trip again at 20 subsequent edits in a day get a stronger warning. At 30 subsequent edits log for Admin investigation. At some point either we have a hound or we have a problematic editor that needs to be cleaned up after. Legacypac (talk) 22:33, 26 June 2017 (UTC)[reply]

idea #2[edit]

Discuss prevention vs intervention[edit]

The AbuseFilter can be used to help editors intervene by alerting to potentially problematic edits, or the software intervene directly. Because in some cases an unsolicited third party intervention could do more harm than good, how do we decide when to add a filter. And which type of filter to use?

Regarding possibly using a disallow filter for 3RR, I don't really support this - it will necessarily have errors and may treat different editing parties in the same dispute differently. — xaosflux ^Talk 02:00, 22 June 2017 (UTC)[reply]
For the most part our communities are operated by consensus, not by "community leaders" - if you mean "administrators" here, just say so. — xaosflux ^Talk 02:01, 22 June 2017 (UTC)[reply]
- I don't mean admins. I've been searching for what to call editors who go beyond just editing articles to participate in policy discussions, dispute resolution, and other aspects of community moderation. I can't find a common term or label. — TBolliger (WMF) (talk) 17:28, 22 June 2017 (UTC)[reply]
  - Wikipedians :D — xaosflux ^Talk 17:36, 22 June 2017 (UTC)[reply]
    - Haha, fair enough. :) Updated! — TBolliger (WMF) (talk) 18:12, 22 June 2017 (UTC)[reply]
      The common term is "editors". :) I boldly made the change. "Wikipedians" is an awkward mouthful, rarely used on-wiki.
      
      I think it's worth examining the fact that we don't have have a word for "editors who go beyond just editing articles". It isn't a very useful concept for us. In fact we're a bit averse to even giving it the status of a term. There are people who happen to specialize in various tasks, but in general anyone can do nearly anything at any time. It is uncommon but possible for someone's first edit to be dispute resolution, tagging an article for deletion, policy discussion, or even directly altering a policy itself. It's quite easy for a first edit to be "content moderation" using the revert button.
      
      Even an admin is "just another editor". They just happen to have additional sensitive buttons, which they are trusted to only push when the community tells them to. (And trusted to judge implicit consensus to use the buttons in routine or urgent cases.) Alsee (talk) 16:14, 25 June 2017 (UTC)[reply]

@Alsee: I don't think it's necessary for us to define this term in any official capacity, but it is important during product design to know your target audiences and personas. There are hundreds of thousands of editors with varying degrees of participation, expertise, and competency. In this case, by saying "community leaders" I meant "user who will go beyond just editing mainspace content and will participate in a discussion about other users." — TBolliger (WMF) (talk) 22:24, 26 June 2017 (UTC)[reply]

General Discussion[edit]

The page says a limitation is "Don't warn users if a filter is set to both warn and disallow." Filter very much can warn first, and only disallow if the editor decides to proceed past the warning. — xaosflux ^Talk 02:05, 22 June 2017 (UTC)[reply]

I agree 100%. I'm going to clarify the bullet on the article page. — TBolliger (WMF) (talk) 17:28, 22 June 2017 (UTC)[reply]

Blank Pages[edit]

Off topic, but I'm not sure how to reach the correct people. There are many blank pages created every day. Some are in Draft space and are not AfC submissions. Some are in Draft or Userspace and are AfC submissions. Some are blank except the AfC starter text or maybe the title repeated or a single link inserted. They all gum things up and waste time. Most are "oh look, what does this submit button do". I believe no editor should be able to create a page without at least say 150 charactors of content. Perhaps a message pops up "please do not make blank or nearly blank pages" with a link to Wikipedia/sandbox. Legacypac (talk) 23:52, 22 June 2017 (UTC)[reply]

@Legacypac: such a limit may be ok for articles but I wouldn't go so far as to say "page". — xaosflux ^Talk 00:22, 23 June 2017 (UTC)[reply]

And it would be WP:VP / WP:EFR to reach "the right people". — xaosflux ^Talk 00:25, 23 June 2017 (UTC)[reply]

Redirects is mainspace would be a valid very short content page. Should be easy to exclude #REDIRECT from the filter. Legacypac (talk) 00:42, 23 June 2017 (UTC)[reply]

@Legacypac: I believe this type of filter should already be create-able, checking for namespace and excluding REDIRECT. Or this could be handled by a bot. This isn't exactly a discussion about Harassment so I think the best place to propose this filter would be Wikipedia:Edit filter/Requested. — TBolliger (WMF) (talk) 17:44, 23 June 2017 (UTC)[reply]

I've moved the discussion. Legacypac (talk) 18:40, 23 June 2017 (UTC)[reply]