Page MenuHomePhabricator

[Spike 1 day] Investigate List Building APIs
Closed, ResolvedPublic2 Estimated Story PointsSpike

Description

Question we are trying to answer

  1. https://reader.wmcloud.org/api/v1/list-reader which looks at similar reader habits,
  2. https://gitlab.wikimedia.org/repos/research/vector-search/ which does a vector search , and
  3. Morelike, which we've already investigated

We should look at (1) and (2) and add rows for
the investigations from this ticket: https://phabricator.wikimedia.org/T369853

I've attached two docs:

  1. A doc I put together with some preliminary findings about these APIs:
  2. A doc Martin put together with additional context:

Acceptance Criteria

Event Timeline

Jdlrobson removed the point value 0 for this task.Jul 29 2024, 6:19 PM
ovasileva renamed this task from [Spike X hours] Investigate List Building APIs to [Spike 1 day] Investigate List Building APIs.Jul 30 2024, 4:22 PM
ovasileva set the point value for this task to 2.

Updated Spreadsheet: https://docs.google.com/spreadsheets/d/1msnUBnfZXV7J2DNs6Py_FkiU2WI_CAo6tFXbpBireF0/edit?gid=1030852928#gid=1030852928
My recommendation:
I would suggest using list building api for the upcoming experiment.
If there is time, the vector search api may be worth a spike, but should be more low priority.

Thank you @KSarabia-WMF, resolving this. We will use the results from this for the UI-less experiment in T371235: Design structure for UI-less experiment comparing API recommendation quality. Adding a small clarification on the experiments - for the search recommendation experiment, we will be proceeding with using MoreLike so that we can draw a more direct comparison to Recent Changes. The API comparison will be used for the UI-less experiment.