Rank eval

When someone runs a search on wellcomecollection.org, we transform their search terms into some structured json. That json forms the query which is run against our data in elasticsearch.
We update the structure of our queries periodically to improve the relevance of our search results.
Every time we update a query, we test it against a set of known search terms, making sure that we're always showing people the right stuff.
You can see the current candidate search queries here.

Precision in works

Queries:

Alternative spellings in works

Queries:

False positives in works

Queries:

  • Maclise

    should not match 'machine'

  • Deptford

    shouldn't match 'dartford' or 'hertford'

  • Sahagún

    shouldn't match 'gahagan'

  • gout

    shouldn't match 'out'

  • L0062541

    shouldn't match 'L0032741' in the title

  • posters

    shouldn't match 'porter'

  • Maori

    shouldn't match 'mary' or 'amoris' or 'maris'

  • monsters

    should not match 'Monastery' or 'Ministers'

Precision in images

Queries:

Recall in images

Queries:

Alternative spellings in images

Queries:

False positives in images

Queries:

  • revolutions

    shouldn't match 'resolutive' or 'Renoult'

  • Maclise

    shouldn't match 'machine'

  • Deptford

    shouldn't match 'dartford' or 'hertford'

  • machine

    shouldn't match 'martin' or 'vaccine'

  • macaronic

  • monsters

    shouldn't match 'Monastery'

  • vestiges

    shouldn't match 'vestitus', 'festival'

  • asylum

    shouldn't match 'slums', 'assumed'

  • maori

    shouldn't match 'mary' or 'mori'