Task 1 — Global Alignment

A traditional/standard OAEI alignment task for DISO-OAEI.

For each pair listed below, you submit a full OAEI alignment between each pair of ontologies. The alignment ontologies are made available within the ontologies directory or can be downloaded as a single zip.

The six alignment ontology pairs

The three jc3iedm-* pairs are coherent under the silver, so repair removes nothing, i.e., $R_{\approx} \equiv R_{\approx}^{+}$.

Reference directories with multiple (supplemental) RDF files

• Rapprox.rdf — $R_{\approx}$ the validated positives (the secondary reference).
• Rplus.rdf — $R_{\approx}^{+} \subseteq R_{\approx}$ the coherence-clean repaired subset; present only for the 3 incoherent pairs.
• incorrect.rdf — human-confirmed true negatives.
• unknown.rdf — mappings whose status remained undetermined after review; the metrics ignore them.

Metric details — partiality-corrected P/R/F1

The silver reference R is partial but locally complete: for every entity it mentions it lists all of that entity’s correct correspondences, but it is silent about unmentioned entities. Let A be your alignment; restrict it to what R can adjudicate:

Submissions are scored using precision, recall and f-measure (under partiality-corrected semantics, as defined above) against their corresponding silver standard reference (download) $R_{\approx}^{+}$ for the primary headline metrics. We also compute the same metrics using their unrepaired variants (included in the same archive) $R_{\approx}$ as secondary (for comparison and reporting purposes only).

Only predictions that are considered un-adjudicable, meaning neither entity is covered by $R$, leave the precision denominator; a prediction over a covered entity that disagrees with $R$ stays in $\hat{A}$ as a genuine error. Equivalence (=) correspondences over named entities are scored — classes, properties, and individuals alike (subsumption cells are ignored). The silver references themselves contain individual–individual equivalences (JC3IEDM country and code individuals; QUDT and OWL-Time units), and these are scored exactly like class correspondences. Self-submission of a reference against itself gives P=R=F1=1.

The leaderboard reports per-matcher micro and macro-averaged precision/recall/f1 across all 6 pairs.

For additional information on local scoring using included tools and the submission format, see this page.