Joint Handling of Semantic Knowledge Resources and their Alignments ?

-

Motivation With the adoption of the Semantic Web vision , semantic knowledge resources (KR), which include taxonomies, structured vocabularies and ontologies, acting as pivotal resources, are nowadays commonly used. This is a direct consequence of the desire to attach formal semantic meaning to manipulated data. These KR, developed by di erent communities with various needs and purposes, are by nature heterogeneous. This heterogeneity leads to the development of systems for nding the correspondences between entities of di erent KR, called alignment [ 2 ]. In addition, KR increasingly involve large volumes of formalized knowledge, containing hundred of thousand entities, which raises the question of generating and validating alignments between large resources.

In order to help nding and reusing these more and more available KRs, a signi cant e ort has been made for providing multi-knowledge resources repositories. This e ort is particularly noticeable in the biomedical domain, where classifying existing objects is a secular tradition.

However, to the best of our knowledge, there is no currently available framework which o ers the possibility to handle both multiple KRs together with their respective alignments, while keeping their native semantics and o ering a support for a transparent visualization of these resources. In addition, with the development of the ontology matching domain, as di erent systems could be used to generate alignments and sometimes relying on user input, either for mappings validation purpose or initial alignment providing, it is a crucial issue to keep track of users and involved alignment methods or tools.

To ll this gap, we have designed the K-Ware framework [ 1 ] which complement existing multi-knowledge resources repositories. Its aim is not to provide a single access point for all available biomedical ontologies and alignments. Rather, it is a framework which could be embedded within projects that have an extensive use of multiple KRs and their respective alignments. In particular, enabling a support for multi point of view navigation and hierarchical visualization of any KR relevant for a dedicated purpose or suitable for a given project. ? This study is supported by the Drugs Systematized Assessment in real-liFe Environment (DrugSafe) platform, funded by the French Agency for Drugs Monitoring.

Joint handling of knowledge resources and alignments 2.1

Alignment management

In order to properly take into account alignments between di erent KRs, we have introduced additional properties to the de nition of the notion of correspondence introduced in [ 2 ]. Therefore, we identify the following components for a correspondence : i) the two entities to be mapped, linked with a MappingRelation, in an directed order (e1-mappingRelation-e2 ); ii) its con dence value; iii) the mapping's author / User; iv) the mapping's method called MappingMethod, either if it is an alignment provided by an automated method ComputerizedProcess with an alignment Tool or a manual one; v) nally, a ag which indicates whether the considered mapping is valid (and validated by a User) for an Alignment between two KRs. 2.2

Handling structural relations within knowledge resources

KRs often exhibits a structural hierarchical organization : the is-a relationship translated into rdfs:subClassOf or the part-whole relationship for formal ontologies and the narrower/broader relationship for taxonomies, translated into skos:narrower and skos:broader respectively. We nd similar notions when it comes to express inter relationships between entities in di erent KRs. For instance, the skos:narrowMatch is used to state a hierarchical mapping link between two conceptual resources in di erent concept schemes. Hierarchical aspect is the the main information to be kept when one wishes to integrate many different semantic resources.

To allow navigating easily KR represented in the OWL or SKOS languages, we distinguish three types of relations : HierarchicalRelation, LiteralDe nition and MappingRelation. In a given KR, a hierarchical relation could be rdfs:subClassOf, skos:broader, part of, etc.). A literal de nition helps rendering a human readable description of an entity (rdfs:label, skos:prefLabel, etc.). While a mapping relation (for instance owl:equivalentTo, skos:exactMatch) handle correspondences between entities.

An API has been implemented for the features de ned models of K-Ware4.

References

1. Thiao-Layel , B. , Jouhet , V. , Diallo , G. K-Ware: vers une gestion conjointe de ressources semantiques et leurs alignements 6iemes Journees Francophone sur les Ontologies , Oct 2016 , Bordeaux, France Hepp, M., de Bruijn , J.

2. Shvaiko , P. , Euzenat , J.: Ontology Matching: State of the Art and Future Challenges . IEEE Transactions on Knowledge and Data Engineering , Volume: 25 , Issue: 1, Jan . 2013 , Page(s): 158 - 176