Updates:

• Spoke with MM and MW from Growth about my analysis of Newcomer Tasks but also the general framework that I use for evaluating recommender systems. MM very much appreciated the framework, in particular a few components;
  • Breaking down a recommender system into stages.
  • Callling out which stages are under whose 'control' -- e.g., this is where we can intervene, this part is on the editor side and we can only support, etc.
  • Relating the analyses back to the pipeline so they understand how equity shifts throughout it.

I will clean up the framework a bit to share more broadly but the broad stages are:

1. Status Quo: Start with all (biased) content
2. Algorithm Design: Filter down to just content "eligible" for recommendation
3. Prioritization: Select individual pieces of content to recommend
4. Impression: Editors see content
5. Click-through: Editor choose whether or not to accept recommendation
6. Edit: Editor does or does not make the edit
7. Impact: What is cumulative effect of edits on content equity dimensions?

Updates:

• Put together a clean version of the slide deck I presented to MM and MW from Growth: https://docs.google.com/presentation/d/1JPhzOiS5Xdsq0-EbN7m6PHalG9Uro-5X5M3egWB4sWk/edit?usp=sharing
• Will be discussing more in-depth with team in the coming weeks too.
• I still would like to clean up the Meta page so it's more approachable, but the slide deck is close to a place where I'd feel more comfortable sharing out.

Updates:

• prepping for discussion with Martin/Miriam about current framework -- what are the big open RQs? Where do we want to go with this work? Agree on the strategy? etc.
• Meta page (https://meta.wikimedia.org/wiki/Research:Prioritization_of_Wikipedia_Articles/Recommendation) is cleaned up to a better state.

Updates: missed the update a while back that I discussed this work with MG and MR. Some good discussion came out of that, especially on the relationship between the Knowledge Gaps work and this work. We have a follow-up next week to discuss more, in particular what it means to have a "representative" sample of content on Wikipedia (sounds simple but quite difficult to define and very important question for a lot of our recommender systems)..

Update: reworking the framing around the slide deck right now based on feedback from MR/MG. Making explicit the choice around who to empower through our recommender systems and the role that different approaches might take. Plans to bring the recommendations back to Product in Q3. Next week I'll summarize the recommendations and then close out this task.