7. Reasoning: facts nobody typed in | Semantic Layer course

Section 1

The axioms were promises

Next section

Section 2

Defined classes: R1 and R2 fire

Next section

Section 3

Three definitions, three named classes

Next section

Section 4

Profiles, engines and rule languages

Next section

Section 5

Watch the triples appear

Next section

Section 6

Continuous reasoning as an event loop

Next section

Section 7

Why is L-5150 a defaulted loan?

Hands on

Derive the default set

35 min

Run an OWL reasoner (owlrl, or ELK via Jena) over the Meridian ontology plus data, produce the inferred named graph, then query the three default classes and see L-2087 in Operational only, L-4590 in Regulatory and Operational, and L-5150 in all three.

make lab-up            # once, starts Postgres + Fuseki + Ontop
make lab lab07-run-reasoner

Full instructions, starter files and the solution are in labs/modules/lab07-run-reasoner/README.md. The lab checks its own work, so you will know if it passed.

Check yourself

Quiz

7 questions, pass mark 70%. Answers are explained either way, so a wrong guess is not wasted.

1 Ingestion asserts only that L-1042 is a mrb:MortgageLoan. After reasoning, L-1042 is also typed as SecuredLoan, Loan, LoanAccount and Account. What produced those four triples?

A nightly ETL job that denormalises loan types Subsumption over the rdfs:subClassOf chain in the T-Box SHACL validation filling in missing types The npa_flag column in the source system

2 Why do inferred triples go into a separate named graph rather than alongside the asserted data? (pick two)

Select all that apply.

Inferred triples use a different RDF syntax that asserted graphs cannot store You must be able to distinguish what the bank was told from what the bank concluded, for audit Changing one axiom invalidates many derived triples, so the inferred graph gets dropped and rebuilt, which is only safe if nothing asserted lives there Named graphs are the only way to make SPARQL queries run faster

3 Under Meridian's three named rules, where does L-2087 (Daniel's personal loan, 47 days past due, no write-off date) fall?

All three: DEF-REG-1.0, DEF-OPS-1.2 and DEF-ACC-2.0 DEF-OPS-1.2 only, the operational arrears definition DEF-REG-1.0 only, the regulatory definition None of the three

4 A migration bug asserts br:BR-KL01 mrb:borrower cust:C-1001. What does the reasoner do?

Rejects the triple because a Branch is not a Loan Infers that BR-KL01 is a mrb:Loan, because rdfs:domain on mrb:borrower is a licence to conclude rather than a constraint Reports a SHACL violation Nothing, since domain axioms are ignored during reasoning

5 Meridian runs ELK on every change event and HermiT only nightly. What is the reasoning behind that split?

HermiT cannot read Turtle files ELK is restricted to the OWL 2 EL profile so it is fast enough for per-event use, while HermiT covers full OWL 2 DL including axioms like owl:inverseOf that ELK structurally cannot handle ELK handles instance data and HermiT handles only the class hierarchy HermiT is a commercial product with per-run licensing

6 Meridian materialises inferred triples with forward chaining rather than deriving them at query time. What is the main cost, and the main reason they accept it?

Cost is query latency; accepted because storage is expensive Cost is storage and pipeline complexity, roughly 41 million asserted triples becoming about 96 million; accepted because the regulatory return must be reproducible from a dated, queryable set of triples Cost is that inferences become less accurate over time; accepted because approximate answers are good enough There is no cost; forward chaining is strictly better

7 Daniel Okafor's APP-9001 has credit score 640 and DTI 0.41. Which classes does the SWRL rule set derive for it?

EligibleApplication only HighRiskApplication only, and it is routed to human review Both EligibleApplication and HighRiskApplication, which is an inconsistency Neither, because 640 and 0.41 are too close to the thresholds