[D-Req-V01-222] Implement text indexing in "dalia" dataset

RDF predicates for text indexing:

• learning resources:
  • dcterms:title
  • dcterms:description
  • schema:keywords
  • schema:name
  • schema:familyName
  • schema:givenName
• communities:
  • dcterms:title
  • dcterms:description

changed the description

mentioned in commit 24bd174b

There is an architectural issue with the data properties of authors (schema:name, schema:familyName and schema:givenName) in property list dt:learningResourceTexts (see dalia.ttl#L75-85) because authors are nested properties of learning resources. As we are using Jena's "one triple equals one document" concept for text indexing, a text search hit in those data properties of authors cannot immediately return the corresponding learning resource. On the other hand, the "one triple equals one document" concept gives more or less consistent scoring across all indexed fields.

Jena's "one document equals one entity" concept doesn't seem to be working anymore and I couldn't find any examples for it.

Abdel and I have been thinking about an efficient query for "return all learning resources that match the text 'NFDI'":

PREFIX dt: <http://dalia.education/text#>
PREFIX ec: <https://github.com/tibonto/educor#>
PREFIX list: <http://jena.apache.org/ARQ/list#>
PREFIX mo: <https://purl.org/ontology/modalia#>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX schema: <https://schema.org/>
PREFIX text: <http://jena.apache.org/text#>

SELECT DISTINCT ?lr
WHERE {
  (?s ?score ?literal ?g ?prop) text:query (dt:learningResourceTexts 'NFDI') .
  OPTIONAL {
    FILTER isBlank(?s) .
    ?s a ?type .
    FILTER(?type = schema:Person || ?type = schema:Organization) .
    ?list list:member ?s . # equivalent to ?list rdf:rest*/rdf:first ?s but much faster
    ?lr schema:author ?list .
  }
  OPTIONAL {
    FILTER(!isBlank(?s)) .
    ?s a ec:EducationalResource .
    BIND(?s AS ?lr) .
  }
  # Communities also have dcterms:title and dcterms:description fields,
  # so these text search matches have to be excluded.
  FILTER NOT EXISTS { ?lr a mo:Community . }
}
ORDER BY DESC(?score)

With the current dataset this executes in about 35 milliseconds.

Now the caveat: "return all learning resources that match the text '*'" - the text search will return all indexed triples. About 380 milliseconds. This is due to the iterations in the schema:author lists.