eyeprolog

Ranking RDF 1.2 annotated claims

Is the bridge open or closed?

The transport authority says the bridge is closed. An anonymous post says it is open. Both statements arrive as RDF 1.2 annotated claims with a confidence and a source. EyeProlog ranks them by confidence times source trust and derives one auditable decision.

Run it in the playground · Source

The question

Should a traveller cross the bridge, avoid it, or check manually?

This is one of a family of RDF 1.2 examples. RDF data arrives as rdf(Subject, Predicate, Object, Graph) facts, as produced by rdf-prolog-interchange, and the decision is written back in the same shape.

The data

rdf(iri('https://example.org/bridge'), iri('https://example.org/status'), iri('https://example.org/closed'), default_graph).
rdf(iri('https://example.org/authorityClaim'), iri('http://www.w3.org/1999/02/22-rdf-syntax-ns#reifies'), triple(iri('https://example.org/bridge'), iri('https://example.org/status'), iri('https://example.org/closed')), default_graph).
rdf(iri('https://example.org/authorityClaim'), iri('https://example.org/assertedBy'), iri('https://example.org/transportAuthority'), default_graph).
rdf(iri('https://example.org/authorityClaim'), iri('https://example.org/confidence'), literal('98', datatype('http://www.w3.org/2001/XMLSchema#integer')), default_graph).

Each status statement is annotated: a reifier points at the statement and carries who asserted it and with what confidence. A separate trustScore rates each source (95 for the authority, 20 for the anonymous post).

Scoring a claim

  rdf_number(Reifier, 'https://example.org/confidence', Confidence),
  rdf_number(iri(SourceIri), 'https://example.org/trustScore', Trust),
  Score is Confidence * Trust.

This is the last part of annotated_status_claim/3, which first finds the statement, its reifier, and its source. The score is confidence times trust: 98 × 95 = 9310 for “closed”, 35 × 20 = 700 for “open”.

Deciding

bridge_decision(
  [claim(Status, Source, score(Score))|_],
  Action,
  Source,
  Score
) :-
  Score >= 8000,
  status_action(Status, Action).
bridge_decision(
  [claim(_Status, _Source, score(Score))|_],
  verify_manually,
  no_trusted_source,
  Score
) :-
  Score < 8000.

Only the top-ranked claim counts, and only if its score reaches 8000. Otherwise the decision is verify_manually: weak evidence does not become a confident answer.

The answer

eyeprolog examples/rdf12-annotated-claims.pl

result_rdf(iri('https://example.org/bridge'), iri('https://example.org/decision'), iri('https://example.org/avoid_bridge'), default_graph).
result_rdf(iri('https://example.org/bridge'), iri('https://example.org/trustedSource'), iri('https://example.org/transportAuthority'), default_graph).
result_rdf(iri('https://example.org/bridge'), iri('https://example.org/score'), literal('9310', datatype('http://www.w3.org/2001/XMLSchema#integer')), default_graph).

These are the first 3 of 9 results; the rest give each claim’s rank, source, and score. Avoid the bridge, on the authority’s word.

The reason

step(bridge_decision([claim(closed, transport_authority, score(9310)), claim(open, anonymous_post, score(700))], avoid_bridge, transport_authority, 9310),
     rule(14),
     ['Status' = closed,
      'Source' = transport_authority,
      'Score' = 9310,
      'Action' = avoid_bridge],
     [9310 >= 8000, status_action(closed, avoid_bridge)]).

The decision rests on the ranked list, the threshold test 9310 >= 8000, and the mapping from closed to avoid_bridge.

The check

condition('C1', resolution, ok, 30).
condition('C5', re_decision, ok, 12).
condition('C6', boundary_consistency, ok, 0).
obligation(collected, theory_scoped, findall(key(A, B) - claim(B, C, score(D)), (annotated_status_claim(B, C, D), A is - D), [key(-9310, closed) - claim(closed, transport_authority, score(9310)), key(-700, open) - claim(open, anonymous_post, score(700))])).
trusted(1).
verdict(checked_with_obligations).

30 steps replay against their clauses and 12 arithmetic or built-in steps are recomputed. One step is an obligation: the claim that findall/3 found all annotated claims. A checker cannot prove that nothing was missed without re-running the search, so it reports this completeness assumption instead of hiding it.

Try it

Lower the authority’s trustScore from 95 to 50. The top score drops to 4900 and the decision becomes verify_manually with noTrustedSource.
Add a third, highly trusted claim that the bridge is open, and predict the new ranking.
Compare with the simpler annotations deck and the scientific evidence graph.

Learn more in The Art of EyeProlog, Chapter 8.

This site is open source. Improve this page.