Page MenuHomePhabricator

[EPIC] Experiment with making MediaSearch usable on non-commons Wikis
Closed, ResolvedPublic

Description

Some wikis, e.g. Portuguese Wikinews, have expressed interest in using MediaSearch as a search interface for their wiki. Extending MediaSearch to other wikis would also allow us to potentially improve the search experience and explore using structured data to improve search results on the Wikipedias, which is a goal of the SDAW grant.

T278984 was a spike to determine if it is possible and what level of effort would be required to make MediaSearch extensible to other wikis. The spike determined that the effort was a promising one, and this epic was created to experiment more thoroughly and see if we can make it work.

If this experiment is successful, additional product work will be necessary to determine whether additional requirements or design are needed to actually implement MediaSearch on another wiki, and whether to move forward with seeking community partnerships or early adopter wikis to do so. That work is out of scope for this experiment epic.

  • How to request quickview data for cross-wiki files?
  • Fix i18n error handling.
  • Enable changes to tab order
  • Enable default selected namespaces

[ ] Identify to users whether image is hosted locally or in shared repository

Event Timeline

CBogen renamed this task from [EPIC] Make MediaSearch usable on non-commons Wiki's to [EPIC] Experiment with making MediaSearch usable on non-commons Wikis.Aug 5 2021, 4:23 PM
CBogen updated the task description. (Show Details)

"...MediaSearch as a search interface for their wiki."

Am I reading this correctly that this is a front end/UI change only? i.e. using MediaSearch's front end, but not it's back end?

"...MediaSearch as a search interface for their wiki."

Am I reading this correctly that this is a front end/UI change only? i.e. using MediaSearch's front end, but not it's back end?

It opens up the possibility of doing both; but this experiment is only about front-end, that's correct. There's no structured data on any wikis besides Commons yet that we could incorporate into a MediaSearch-based backend. Those types of experiments would have to come later, once we've done more SDAW work to create structured data on Wikipedias.

@Seddon is this correct?

Right now this experiment is focused around the front end, the image metadata that gets surface in extmetadata and how some of that information gets presented to the user. Using the mediasearch search profile when searching for files is a possibility but that is a step further down the road.

Just a note if we move from an experiment to a production implementation: Machine Learning Ranking (MLR) relies on a model of what results user select. This require the proper instrumentation, but also the ability to assign a value to the click (in a list of result, clicking on the third link has more weight than clicking on the first, as it requires more effort). We don't have a good way to have a similar model for a result grid. For Commons, this is a minor issue as we don't do MLR there. For wikis where we do MLR, it might degrade our ability to retrain models. There might be easy workarounds (just drop the data coming from grid results, or similar). This is something that won't break anything right away, but which could have significant impact in the future. We should at least make a conscious decision.

This comment was removed by egardner.

Change 727019 had a related patch set uploaded (by Seddon; author: Seddon):

[mediawiki/extensions/MediaSearch@master] WIP Enable Changes to MS tab order

https://gerrit.wikimedia.org/r/727019

Edu added a subscriber: Edu.

@CBogen the patches for this are merged ... what's the next step?

CBogen claimed this task.

@CBogen the patches for this are merged ... what's the next step?

I think we can close this out!