Page MenuHomePhabricator

Add articletopic model to testwiki
Open, Needs TriagePublic

Description

We are aiming to use testwiki as an introduction to Growth features (see parent task). It would be nice if those features could include ORES topics. If the articletopic model would be enabled on that wiki, I think the rest would just work (after a Cirrus refresh). The quality of predictions doesn't matter much, so the enwiki model could be reused.

If it's not much work, please enable it.

Event Timeline

Tgr created this task.Jul 6 2020, 7:37 PM
Restricted Application added a project: Growth-Team. · View Herald TranscriptJul 6 2020, 7:37 PM
Restricted Application added a subscriber: Aklapper. · View Herald Transcript

@Tgr -- which team would be doing this work? Is it Growth, Scoring, or Search? Or all three?

Halfak added a subscriber: Halfak.Jul 7 2020, 8:49 PM

Aha! We don't have it here, but we could. See https://ores.wikimedia.org/v3/scores/testwiki/

I'm thinking that we would just re-apply the enwiki topic model in testwiki.

Tgr moved this task from Inbox to External on the Growth-Team board.Jul 7 2020, 10:36 PM

@Tgr -- which team would be doing this work? Is it Growth, Scoring, or Search? Or all three?

This is work for the Scoring team (a simple configuration change, hopefully).

I'm not 100% sure if the Search part of the pipeline would work out automagically, but I think it would.

I'm thinking that we would just re-apply the enwiki topic model in testwiki.

That would be ideal, the text of the articles is taken from enwiki so it would probably give reasonable predictions. Although even if it doesn't, having the topic filter feature available would be a plus, even if the results are random.

Thanks for helping on this, @Halfak. Is this something you can easily knock out, or should we plan it for later?

@EBernhardson @dcausse -- do you know if the pipeline part would just work?

EBernhardson added a comment.EditedJul 9 2020, 12:24 AM

For the elastic update pipeline, once mediawiki starts reporting scores for testwiki to mediawiki_revision_score events, and the ores api reports testwiki as having the appropriate model (for fetching thresholds), then everything else should "just work". If we want to populate the full set of pages, rather then only the ones with revisions after the deployment date there is a small amount of work to do, but it shouldn't be too complex.

Change 610851 had a related patch set uploaded (by Halfak; owner: Halfak):
[mediawiki/services/ores/deploy@master] Adds drafttopic and articletopic models for testwiki

https://gerrit.wikimedia.org/r/610851

Halfak added a comment.Jul 9 2020, 3:12 PM

It's a pretty simple config change from our end. No big deal.

Thanks, @Halfak!

Yes, @EBernhardson, we will want to score pages with revisions before the deployment (the pages that already existed on Test Wikipedia).

Change 614834 had a related patch set uploaded (by Halfak; owner: Halfak):
[mediawiki/services/ores/deploy@master] Adds draft/articletopic models to testwiki

https://gerrit.wikimedia.org/r/614834

Change 610851 abandoned by Halfak:
[mediawiki/services/ores/deploy@master] Adds drafttopic and articletopic models for testwiki

Reason:

https://gerrit.wikimedia.org/r/610851

Change 614834 merged by Calbon:
[mediawiki/services/ores/deploy@master] Adds draft/articletopic models to testwiki

https://gerrit.wikimedia.org/r/614834