Create a new search profile in WikibaseMediaInfo that uses the new weighted_tags data added in T286562
Probably the best approach will be a similar way to how we do a statement_keywords search atm - i.e. query wikidata for matches for the search string, and then add a bunch of Match queries against weighted_tags using the returned item ids
The profile should be triggerable via a url param
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Add weighted_tags queries for mediasearch | mediawiki/extensions/WikibaseMediaInfo | master | +828 -70 |
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | CBogen | T299781 [EPIC] Image suggestions backend | |||
| Resolved | CBogen | T281582 [EPIC] Develop a confidence score for MediaSearch results | |||
| Resolved | Cparle | T283869 [EPIC] Experiment with incorporating Image Matching Algorithm data into commons search index | |||
| Resolved | Cparle | T286563 [M] Create new search profile for commons that uses weighted_tags |
Change 732956 had a related patch set uploaded (by Cparle; author: Cparle):
[mediawiki/extensions/WikibaseMediaInfo@master] Add weighted queries for mediasearch
Note that the score returned from this query can be used as a confidence score. Here are the scores for the labeled data we have. Scores below 40 tend to overestimate the likelihood that an image is good, but it's unlikely anyone will be interested seeing images where we think there's a <40% chance of it being a good match, so IMO this confidence score is usable
| Score range | proportion of images in score range that are good |
| 0-20 | 0.0704 |
| 20-30 | 0.1673 |
| 30-40 | 0.2635 |
| 40-50 | 0.4068 |
| 50-60 | 0.5278 |
| 60-70 | 0.6628 |
| 70-80 | 0.7358 |
| 80-90 | 0.8989 |
| 90-100 | 0.9512 |
Change 732956 merged by jenkins-bot:
[mediawiki/extensions/WikibaseMediaInfo@master] Add weighted_tags queries for mediasearch