This paper presents the results from the CANARD system in the OAEI 2020 campaign. CANARD is a system able to generate complex alignments. It is based on the notion of competency questions for alignment, as a way of expressing user needs. The system has participated in tracks where instances are available (Populated Conference, Populated Geolink, Populated Enslaved and Taxon datasets). This is the third participation of CANARD in the OAEI campaigns.
1.1
Following the evaluation made in [
Automatic generation of CQAs OAEI tracks do not cover CQAs i.e., the
CQAs can not be given as input in the evaluation. We extended last year's query
generator so that it can output binary queries. The query generator now produces
three types of SPARQL queries: Classes, Properties and Property-Value pairs.
Classes For each owl:Class populated with at least one instance, a SPARQL
query is created to retrieve all the instances of this class. If <o1#class1> is a
populated class of the source ontology, the following query is created:
SELECT DISTINCT ?x WHERE f?x a <o1#class1>.g
Properties For each owl:ObjectProperty or owl:Dataproperty with at least one
instantiation in the source knowledge base, a SPARQL query is created to
retrieve all instantiations of this property. If <o1#property1> is an instantiated
property of the source ontology, the following query is created:
SELECT DISTINCT ?x ?y WHERE f?x <o1#property1> ?y.g
Property-Value pairs Inspired by the approaches of [
Implementation adaptations In the initial version of the system, Fuseki server endpoints are given as input. For the SEALS evaluation, we embedded a Fuseki server inside the matcher. The ontologies are downloaded from the SEALS repository, then uploaded in the embedded Fuseki server before the matching process can start. This downloading-uploading phase takes time, in particular when dealing with large les.
The CANARD system in the SEALS package is available at http://doi. org/10.6084/m9.figshare.7159760.v2. The generated alignments over the datasets in which CANARD performed are available at: { Populated Conference: http://oaei.ontologymatching.org/2020/results/ complex/popconf/results_conference.zip { Populated GeoLink: http://oaei.ontologymatching.org/2020/results/ complex/popgeolink/popgeolink_results_2020.zip { Populated Enslaved: http://oaei.ontologymatching.org/2020/results/ complex/popenslaved/popenslaved_results_2020.zip { Taxon: http://oaei.ontologymatching.org/2020/results/complex/taxon/ results_taxon.zip 2
Please refer to http://oaei.ontologymatching.org/2020/results/complex for the results of CANARD in the OAEI 2020 campaign.
Two datasets were used in the Populated Conference subtrack, one with more instances than the other. CANARD could perform all the matching tasks in the smaller dataset but timed out on 16 out of the 20 oriented pairs. This highlights one of CANARD's limitations : scalability.
For this reason, the coverage score is much lower on the large dataset than on the small one. While merging the results of all matchers by taking their best run (original, small or large dataset), CANARD obtains the best coverage score. It is the only evaluated matcher with a Coverage score higher than that of the reference simple alignment (ra1).
CANARD achieved the best relaxed-precision score (0.89) and the second best relaxed-recall score (0.54). This score however does not consider the semantics of the output correspondence. Most systems achieved a high relaxed precision score. Because of the automatic generation of CQAs, many correspondences of the form 9gbo:hasPlatformType.fXg gmo:Platform, where X is a platform type were found.
CANARD performed the lowest in this track out of the three evaluated complex matchers. On the enslaved-wikidata oriented pair of ontologies, CANARD found many instance links for each support answer. These links were found with literal comparison on two instances, a generic method which brings a lot of errors on a dataset with many literal information (such as dates or values). In the case of binary CQAs, as CANARD tries to nd a property path between each aligned entity, the runtime exploded and had to be stopped. This shows a major aw in CANARD that should be xed. 2.4
CANARD has output much more correspondences than last year. A recurring pattern was found in the correspondences: an object property from Taxref or Agrovoc is aligned to a chain of agronomicTaxon:hasHigherRank and agronomicTaxon:hasLowerRank properties. This lowered the precision score in comparison with last year's. This can be explained by: { Wrong instance linking based on label matching regardless of the language (e.g., a plant taxon matched to a habitat) { The computation of all possible links between the two matched instances in the target knowledge-base { If a path is found between two matched instances, it gets a default con dence value of 0.5. If a better path is found (a path with a lexical similarity to the source property), it gets a higher value and the default path are ltered. In this case, no better path was found so all correspondences were kept.
The recall score is also lower as last year's because the system was set to use only 2 support instances this year. As the instances are not homogeneously described in each dataset, more support instances mean more chances of nding one in the target dataset which instantiate the initial knowledge need. However, CANARD still achieves the best Coverage scores. 3
CANARD relies on common instances between the ontologies. It works best with aligned instances as it will try to nd lexically similar entities otherwise. Hence, when such instances are not available, the approach is not able to generated complex correspondences. Furthermore, CANARD is need-oriented and requires a set competency questions to guide the matching process. Here, these \questions" have been automatically generated based on a set of patterns.
CANARD's runtime is extremely long. It depends (among other things) on the performance of the SPARQL endpoint it interrogates and the presence (or not) of equivalent links.
However, even with generated queries (instead of user input CQAs) it obtains some of the best coverage scores. 4
This paper presented the adapted version of the CANARD system and its preliminary results in the OAEI 2020 campaign. This year, we have been participated in Populated Conference, Populated GeoLink, Populated Enslaved and Taxon track, in which ontologies are populated with common instances.