The Edit filter is a tool used to allow trusted users to set specific controls on user activity and create automated reactions for certain behaviors.
The Abuse Filter extension was developed by Werdna with support from the Wikimedia Foundation, and went live on the English Wikipedia in March 2009. As of July 2009, nearly all user-facing elements of the filter refer to it as the "Edit filter" as not all the edits it flags are abusive. In the software and special pages it retains its original name.
The extension allows automatic filters/heuristics to be applied to all edits. Specific rules can be developed, such as "users with fewer than 500 edits are blocked from moving pages to titles which match this regular expression: /poop/". Of course, the rules can get quite a bit more complicated.
Log entries are viewable by all users, and while filters are by default publicly viewable, others are set to be private. For all filters, including those hidden from public view, a brief, general summary of what the rule targets will be available, and displayed in the log, the list of active filters, and in any error messages generated by the filter.
Implementation of AbuseFilter on Wikipedia is being done with due caution—most abuse filters should be tested for a few days (in "log only" mode) before being brought to full force ("warn", "disallow" or "throttle" modes). Only members of specific groups are allowed to modify any of the filters, and this group is assignable by administrators.
Presently, requests for assignment of the "Edit Filter managers" group to non-admins should be made at Wikipedia talk:Edit filter, where a discussion will be held for up to a week prior to a decision being made.[1]
Extension documentation
 | The below was written when the extension was first written, and may be somewhat out of date. |
Filtering criteria
For all of the following, we can do extensive normalisation, regex matching, length comparison and regular comparisons (less than, greater than, equal to) matching, combining different filters with boolean logic.
User
- Edit count.
- Account age.
- Groups.
- Email-confirmed status.
Titles (moved-to, moved from included)
- Namespace.
- Title.
- Full text.
- Restrictions and protection status.
Action
- The action type (edit, move or createaccount).
- Edit summary.
- Contents of the edit.
Throttling
- Filters can specify whether actions done at a certain rate are by the same IP address, account, /16 range, account-creation-date, and/or to the same page, for a consequence (below) to be invoked.
- Any of the above conditions can be combined to produce a separate rate-limiter. For instance, we can group all accounts created on the same date, from the same /16, for the purposes of rate-limiting.
- Any actions set for that filter will occur if, and only if, the rate-limiter is tripped. This reduces false-positives by making the filter apply only if the same user is consistently tripping a particular filter, rather than a single false-positive.
Actions which can be assigned in response to filtered edits
If a user triggers a filter, AbuseFilter can apply any of the following sanctions based on the severity of the offense:
- All actions triggering a filter are logged at a special page.
- The user's action can be tagged for further review.
- The user can be warned that their actions may be unconstructive.
- The user's action may be disallowed.
- The user's account may have its autoconfirmed status removed
The following actions are currently not available on this wiki:
- The user's account may be blocked from editing, along with all IP addresses used in the last 7 days.
- The user's account may be removed from all privileged groups (such as sysop, bot, rollbacker).
Note: Individual sanctions can be disabled selectively. Any edit filter manager can restore autoconfirmed status in case of an error.
Monitoring
All edits triggering an action will produce a report at Special:AbuseLog. On this page, a brief log entry is entered. Users with the appropriate permissions may view the log summary. Users with certain higher permissions may view details on the log entry. This includes all information available to the filter when it ran, and may be useful for debugging purposes. Users with the highest level of log-viewing permissions may view private data about the action which caused the log event, such as the user's IP address. See the AbuseFilter documentation for more details on the permissions structure.
Sample abuse log entries
- 06:43, 23 June 2008: Andrew (Talk | contribs | block) triggered an abuse filter, making an edit on Main Page. Actions taken: warn,disallow; Filter description: Test Filter
- 06:43, 23 June 2008: Andrew (Talk | contribs | block) triggered an abuse filter, making an edit on Main Page. Actions taken: none; Filter description: Test Filter
Sample detailed abuse log entries
A sample detailed log entry
- 06:43, 23 June 2008: Andrew (Talk | contribs | block) triggered filter 1, making an edit on Main Page. Actions taken: warn,disallow; Filter description: Test Filter (details)
- 06:43, 23 June 2008: Andrew (Talk | contribs | block) triggered filter 2, making an edit on Main Page. Actions taken: none; Filter description: Test Filter (details)
- 06:42, 23 June 2008: Andrew (Talk | contribs | block) triggered filter 1, making an edit on Main Page. Actions taken: warn; Filter description: Test Filter (details)
- 06:42, 23 June 2008: Andrew (Talk | contribs | block) triggered filter 2, making an edit on Main Page. Actions taken: none; Filter description: Test Filter (details)
- 06:22, 23 June 2008: Andrew (Talk | contribs | block) triggered filter 1, making an edit on Main Page. Actions taken: warn,disallow; Filter description: Test Filter (details)
- 06:22, 23 June 2008: Andrew (Talk | contribs | block) triggered filter 2, making an edit on Main Page. Actions taken: none; Filter description: Test Filter (details)
The details link brings up a screen like that on the right.
Safeguards
To protect the wiki against poorly configured filters, a technical limit is imposed on the maximum percentage of actions that will trigger a given filter. Other technical limits are in the process of being written.
Notification
All notifications are based on the template {{edit filter warning}}.
Standard notifications shown to a user triggering a filter action:
| Message name | Message text |
| abusefilter-disallowed |  | An automated filter has identified this edit as potentially unconstructive, and it has been disallowed. If this edit is constructive, please report this error. | |
| abusefilter-degrouped |  | An automated filter has identified this edit as potentially unconstructive, and it has been disallowed. In addition, your account has been removed from privileged groups. If this edit is constructive, please report this error. | |
| abusefilter-autopromote-blocked | | An automated filter has identified this edit as potentially unconstructive, and it has been disallowed. In addition, your account has been temporarily restricted from executing some sensitive operations. If this edit is constructive, please report this error. | |
Generic warning message is below. Admins are advised to use custom warnings.
Some existing filters and their warnings:
| Filter and message | Message text |
3: blanking articles
blanking |  | Warning: An automated filter has identified this edit as potentially unconstructive. Please do not replace Wikipedia pages with blank content. Blank pages are harmful to Wikipedia because they have a tendency to confuse readers. - If it is a duplicate article, please redirect it to an appropriate existing page.
- If the page has been vandalised, please revert it to the last legitimate version.
- If you feel that the content of a page is inappropriate, please edit the page and replace it with appropriate content or leave a message on the talk page.
- If you believe the page should be deleted altogether, please see the deletion policy for information on how to proceed.
- If this edit is constructive, please click 'Save page' again, and report this error.
| |
30: large deletions
removal |  | Warning: An automated filter has identified this edit as removing a large amount of content, and potentially being unconstructive. Some material on Wikipedia should be removed, for example vandalism and false or libelous information. If this is what you're doing, and you believe your removal is correct, please go to the bottom of this page and press 'Save page' to continue. If not, and your edit is clearly not constructive, you may be blocked from editing Wikipedia. If you did not remove a large amount of content, and this message is in error, please report this. | |
18: test edits
testeditwarn | | Welcome to Wikipedia! Your edit has triggered an automated filter because it looks like you're just testing out editing. If you were intending to make a test edit, try using our sandbox instead — that's a page that is specifically set aside for people to work out how to use wikitext. Please take some time to look over Help:Editing for more instructions on how to use wikimarkup, and feel free to ask at the Help Desk if you have any questions. If you're sure you still want to make this edit, go to the bottom of this page and click 'Save page' again, and report this error. | |
If a filter is set to warn and disallow, then a user clicking "Save page" will alternatively see that warning and standard disallowed message.
Known issues
When the extension is initially installed, the available actions will not include blocking or removing from privileged groups. This restricted usage has been determined by community consensus, and if the extension is successful, the community may decide to enable the block, rangeblock or degroup actions for use on this wiki. Admins wishing to block users for triggering the abuse filter can use the {{uw-afblock}} template.
The full technical details of implementation are available on the bug (bug 15684).
Notes
- ^ Process informally agreed upon when Cobi (talk · contribs), who was not yet an admin at the time, was granted the userright in this thread: Wikipedia talk:Edit filter/Archive 3#Abuse Filter editors group membership request.
