Tasks Overview

Task 1 — Global alignment

The standard OAEI alignment task of DISO-OAEI: for each of the six ontology pairs below, you submit one full OAEI alignment.

The DISO-OAEI alignment ontologies are made available in the ontologies directory, or they can be downloaded conveniently as a zipped archive. The six alignment ontology task pairs are broken down in further detail below.

Submissions are scored using precision, recall and f-measure (under partiality-corrected semantics) using their corresponding repaired silver standard reference alignment. Both $R_{\approx}$ and $R_{\approx}^{+}$ are given; submissions are headlined against $R_{\approx}^{+}$ (scores against $R_{\approx}$ are also reported). Results are both micro (and macro) averaged across all six task pairs when published to the baselines page (and, for participant submissions, the live leaderboard).

Download the ontologies and the reference alignments for Task 1; and see Task 1’s extended documentation, its submission format, and the quickstart for submission logistics.

Task 2 — Local equivalence ranking (mixed-signature, with NIL mappings)

Task 2 is an equivalence ranking task, for which a brief overview is provided in the table below.

Task Definition: Given a pair of ontologies, $\mathcal{O}_{src}$ and $\mathcal{O}_{tgt}$, a source entity from $\mathcal{O}_{src}$, denoted $e_{src}$, a set of 50 potential target entities (consisting of 49 entities from $\mathcal{O}_{tgt}$ and a NIL entity), denoted $\mathcal{K}$, produce a ranking for $\mathcal{K}$, ordered by confidence that $e_{src} \equiv e_{tgt}$ for each $e_{tgt}$ in $\mathcal{K}$. In the case where an equivalence correspondence between $\mathcal{O}_{src}$ and $\mathcal{O}_{tgt}$ through $e_{src}$ does not exist, instead map $e_{src}$ to NIL.

Simply put: for each source entity ($e_{src}$) you are given a fixed candidate pool of 50 possible target entities (represented by their IRI; $e_{tgt}^{1}, e_{tgt}^{2}, \ldots, e_{tgt}^{50}$) drawn from the target ontology’s signature. A participating system aims to rank order the pool such that the true equivalence correspondence $\langle e_{src}, e_{tgt}^{i}, \equiv, [0,1] \rangle$ appears as early as possible; or, when the source has no equivalent target from $\mathcal{O}_{tgt}$ within the pool, your ranking should place the NIL IRI as early as possible.

Candidates: are pooled into one unified set per source entity, each pool has a cardinality (or size) of 50; that is, 49 + a NIL candidate. The NIL candidate (represented by the IRI https://oaei.ontologymatching.org/2026/diso/NIL) indicates that a mapping does not exist from the source entity to any other possible candidate entity. Read the justification for how these are computed here. Given that the query set is relatively small (290 queries: 217 uco-stix + 73 stix-d3fend), it remains infeasible to provide a public training set. As such, this task is considered unsupervised; however, there are no formal constraints on the methodology used to implement a participating system, so long as the system is documented and the results are reproducible.

Download the candidates for Task 2; and see Task 2’s extended documentation, its submission format, and the quickstart for submission logistics.