The Research team in collaboration with the ML-Platform team are creating a new service to help patrollers to detect revisions that might be reverted.
Requirments:
Please follow the progress of this project on the related tasks.
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| In Progress | diego | T314384 Develop a ML-based service to predict reverts on Wikipedia(s) | |||
| Resolved | diego | T314385 Create a language agnostic model to predict reverts on Wikipedia | |||
| Resolved | diego | T314386 Create a multilingual model to predict reverts on Wikipedia | |||
| Duplicate | None | T329071 Integration of Revert Risk Scores to Recent Changes as a filter | |||
| Resolved | diego | T336421 Revert prediction data request | |||
| Open | MunizaA | T343061 Denylist for language agnostic revert risk model | |||
| Open | fkaelin | T349755 Training pipeline for Revert Risk Language Agnostic (RRLA) model | |||
| Resolved | XiaoXiao-WMF | T351897 Set the thresholds Revert Risk models to be used on the Recent Changes Feed (via ORES Extension) |
It has been decided to focus on knowledge integrity risks from two categories of our taxonomy:
For the records here a snippet (by @achou) to try the models from the WMF's cluster
Language-Agnostic:
curl "https://inference.svc.codfw.wmnet:30443/v1/models/revert-risk-model:predict" -d @input.json -H "Host: revert-risk-model.experimental.wikimedia.org" --http1.1 -k
Multilingual:
curl "https://inference-staging.svc.codfw.wmnet:30443/v1/models/revert-risk-model:predict" -d @input.json -H "Host: revert-risk-model.experimental.wikimedia.org" --http1.1 -k
An example for input.json: { "lang": "ru", "rev_id": 123855516 }
What does the timeline/roadmap look like for getting this model into Liftwing and available as an API? Our team is considering working on a project next year leveraging this model and it would be helpful to know what the timeline would be.
cc @calbon
I've talked with the ML-Platform folks, and we are going to have this API available within this month. @achou is working on this, and will let us know when the public end-point is available
Both models (Language-Agnostic and Multilingual) have been deployed to Lift Wing production. (T332998, T333124) The next step is to work on the public endpoints.
@Samwalton9, will your team be using internal endpoints or public endpoints for the project? Is there any more documentation about this potential project?
Thanks for the update @achou! You can find further information on our plans so far at T336934 or https://docs.google.com/presentation/d/1YiF9rfDKoTvoKVdRUYXAXB2jl6oQhn5-55emQ6TQCg8/edit. We're still in the very early phase of planning on this so I don't have a good answer for you about internal vs public just yet.