Page MenuHomePhabricator

Investigate training ORES with AbuseFilter conditions
Open, HighPublic

Description

AbuseFilter contains 192 filters on English Wikipedia.

These filters can tell us:

  1. What edits constitute as abuse.
  2. What was used to determine that they are abuse.
  3. The severity of the abuse.

We should investigate using this data to better train ORES to detect vandalism.

With this data, perhaps it would be possible for ORES to be better at identifying what is vandalism and with any hope exceed the reliability of AbuseFilter at identifying and preventing abuse.

Event Timeline

dbarratt created this task.Feb 1 2018, 7:44 PM
Restricted Application added a project: Scoring-platform-team. · View Herald TranscriptFeb 1 2018, 7:44 PM
Restricted Application added a subscriber: Aklapper. · View Herald Transcript
TBolliger updated the task description. (Show Details)Feb 1 2018, 7:48 PM
TBolliger added a subscriber: TBolliger.
He7d3r added a subscriber: He7d3r.Feb 2 2018, 9:59 AM

I agree, the idea is good, though we would need some ways to set a "quality" for each filter, e.g. T30213.

Harej triaged this task as High priority.Apr 9 2019, 9:21 PM