Aaron Halfaker asked to write this down as a ticket in Wikimedia Hackathon 2017.
Create a model which tries to determine if an edit will be manually approved with FlaggedRevs OR if it will be reverted. User interface isn’t required except for API support. The goal is to see how well a model which has been taught using Flagged Revs approves and reverts works compared to goodfaith and damaging filter.
Good diffs (approved manually)
- This is too slow for Quarry, but fast in labs console.
- In a single approved diff there can be one or more edits and editors
Bad diffs (changes generated using reverts detected by summary)