Test Autocomplete

Perform a test where we search for important people/artists/objectTypes (the italic category below), and hope they will come up amongst the top 10-20.

• Start with a prefix of 4 letters (eg "brit", "remb" "new y"). If the term is not in the top 20, increase to 5 and 6 letters.
• Record the prefix length and the rank
• If an "unimportant" term (eg "Paris, Texas") ranks above the important term (eg "Paris France"), record this

Coreferenced terms (having 2 URIs) are marked in bold, since they may have some special behavior
Mitac has implemented this as an automated test (AutocompleteTest.testRDFRank()), see results below.

Test results from the backend:

Test results from the UI (here Rank signifies the place in the autossugested list):

Test RDFRank

The following queries showed that RDFRank has reasonable relevance on the data we have (tested on 115k BM objects).

• But they don't involve a Lucene search, so to be sure we also did Test Autocomplete.
• RDFRank is only 2 decimals and the spread is not so large. But still, it has sufficient discriminative power for 50-100 results.
  (we don't need to investigate histograms, just top 100)

Places

PREFIX rank: <http://www.ontotext.com/owlim/RDFRank#>
PREFIX skos:  <http://www.w3.org/2004/02/skos/core#> 
SELECT ?place ?label ?rank WHERE {
  ?place skos:inScheme <http://collection.britishmuseum.org/id/place>;
         rank:hasRDFRank ?rank;
         skos:prefLabel ?label .
} ORDER BY DESC(?rank) LIMIT 100

It returns first the continents and large regions (eg British Isles) because they have a lot of subplaces (P88_consists_of, P89i_contains) and rso:FR7_from_place inferred through the place hierarchy.

Cities

Now let's see which are the most popular "town or city" (K) or "village or settlement" (O)

SELECT ?place ?label ?rank WHERE {
  ?place skos:inScheme <http://collection.britishmuseum.org/id/place>;
         crm:P2_has_type ?type;
         rank:hasRDFRank ?rank;
         skos:prefLabel ?label .
  FILTER (?type=<http://collection.britishmuseum.org/id/place/type/K> ||
          ?type=<http://collection.britishmuseum.org/id/thesauri/place/type/O>)
} ORDER BY DESC(?rank) LIMIT 100

Some of the important cities indeed come first.
Some are a bit curious (Samarkand, Quirigua) but I checked Samarkand and indeed we have about 100 objects from there, and depicting it (P62_depicts)

Organizations and People

PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
PREFIX rank: <http://www.ontotext.com/owlim/RDFRank#>
PREFIX skos:  <http://www.w3.org/2004/02/skos/core#> 
SELECT ?agent ?label ?rank WHERE {
  ?agent skos:inScheme <http://collection.britishmuseum.org/id/person-institution>;
         rank:hasRDFRank ?rank;
         rdfs:label ?label .
} ORDER BY DESC(?rank) LIMIT 100

The BM comes up first. Hurray!

People

Only people:

PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
PREFIX rank: <http://www.ontotext.com/owlim/RDFRank#>
PREFIX skos:  <http://www.w3.org/2004/02/skos/core#> 
PREFIX crm:   <http://erlangen-crm.org/current/> 
SELECT ?person ?label ?rank WHERE {
  ?person skos:inScheme <http://collection.britishmuseum.org/id/person-institution>;
         a crm:E21_Person;
         rank:hasRDFRank ?rank;
         rdfs:label ?label .
} ORDER BY DESC(?rank) LIMIT 100

Counts - temp table by Mitac