| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Declined | • Deskana | T113379 EPIC: Investigate adding aliases from Wikidata into the search index so that they can enhance the results and reduce the zero results rate | |||
| Resolved | • Deskana | T114867 Evaluate the benefits of adding wikidata aliases to cirrus indices |
Event Timeline
According to T114867 this does not seem to be worth a try.
We should maybe look at another solution. Issuing a query against wikidata seems to be a better idea.
This feature is enabled on itwiki, this would cover the following use case :
- query with an alias missing in the wiki index
- query with a name in a different language
- query with small transliteration variations
Problem with the current wikidata integration is that it's not optimal, results from wikidata should be integrated in a seamless manner into the search results. Moreover if the wikidata entity has a link to user wiki.
This involves :
- UI design/work
- maybe some backend work (performance considerations: I'm not sure that our cluster will support doubling the number of queries)
After having a look at how we index wikidata content it looks like it's not very optimal. Structured data seems to be flattened to the content field (https://www.wikidata.org/wiki/Q216092?action=cirrusDump). I've heard that the wikidata team plans to use cirrus prefixsearch for wbsearchentities. We could maybe start thinking on what could be a good mapping for wikidata docs?
The analysis performed in T114867: Evaluate the benefits of adding wikidata aliases to cirrus indices indicated that this avenue is not worth pursuing at present. Therefore, I am declining this epic. It can be reopened if we decide to pursue this in the future.