This paper describes ONTMAT an ontology matching system, and presents the results obtained for the Ontology Alignment Evaluation Initiative (OAEI) 2017. ONTMAT is an ontology matching process, which compares the instances of ontologies to align in order to deduce the relations between their concepts. Then, based on hierarchical and binary relations between the concepts inside the ontologies it performs entities matching.
Presentation of the system ONTMAT (ONTology MATching) is an ontology alignment tool, aiming to align OWL entities (classes, object properties i.e. binary relations), participating for the first time in OAEI (Conference track).
ONTMAT uses a terminological methods based on WordNet dictionary [2],which is exploited as background knowledge to provide a set of the relations between individuals names of the ontologies source ( ) and target( ). Then, if the name does not exist in WordNet the approach handles the n-gram measure instead of the dictionary. Moreover, from this set of individual relations we will deduce the equivalence or subsumption relation among their concepts. The equivalent concepts are recorded in an alignment matrix (AM), and the concepts related by subsumptions relations are registered within a temporary alignment matrix (TAM) [4]. Furthermore, the TAM elements and the concepts neighbors of AM are compared by using the inference roles with the terminological techniques cited previously and the retained alignment will be added to AM. The concepts neighbors are those related by hierarchical or binary relations with AM concepts. Here, we first align the neighboring concepts because they have more chance to be similar [1], after we will align the other concepts by using the same technics. Next, inference technics are applied on AM to align the binary relations.
In our proposition we suppose that Wordnet is hierarchically organized as W(S,≤,A g ,g), where S is a set of synsets {s1,s2,…,si}(i is a positive integer), and an annotate function A g associates the gloss g to each synset. Furthermore, the relations ≤ between concepts , may be presented in the following logical relations [4] as: 1)
;means that is a hyponym or meronym of ; 2)
; express that is a hypernym or holonym of ; 3)
; signified that and belong to the same synset are similar [1]; 4) when and are the siblings in the part of hierarchy they are connected by a relation of antonymy.
The entities aligned can be related by one of the hierarchical relation presented in the set , , where ( : equivalence; subclass), fuzzy relation symbolized by "&", or binary relation. Further, the binary ontologies relations (O 1 , O 2 )are also aligned by an element of the set HR. The algorithm will explain in the following items:
1. In level 1 we compare the instances names (I O1 ,I O2 ) of ontologies( , )to deduce the relations among their concepts. To do this, WordNet is exploited because we cannot assume with certainty that two entities are dissimilar if they have different names (synonyms), or they are equivalent if they have the same name(homonyms). If the name does not existed in WordNet we will measure the similarity among names by the n-gram measure. Then the equivalent instances will construct the instances matrix IM. The concepts (C 1 ,C 2 ) of ( , ) that have the same sets of instances in IM are considered as equivalent concepts as proven in [4], and ( , , ) can be added to AM. Although, if the instances set of C 1 are included in the instances set of C 2 then ( , , ) will be inserted in TAM.
Next the results obtained will be combined with inferences methods illustrated in [4] to be inserted in AM: ( , , ) of TAM confirmed or modified, where , , , . 3. The concepts neighbors sorted by hierarchical relations of the AM elements ( , , ), are ( , linked to ( , respectively by an element of the set HR. The neighbors , joined by " " with , of AM in ( , ), will be aligned by using the inferences techniques applied on the background knowledge [3]. The background knowledge is the ontology source when the neighbors belong to O1, and ontology target if we match the neighbors existing in O2. The other neighbors will be matched using the terminological methods. 4. The fourth level exploits the description logic roles proven in [4] to match the concepts ( , associated to ( , by binary relations ( , ) in ( , ), as following: ─ If ( , , ( , , then ( , , will be inserted to A BRS M (Alignment Binary Relation Source Matrix) ─ If ( , , ( , , then ( , , will be added to A BRT M (Alignment Binary Relation Target Matrix) Thus, binary relations can be aligned because we have: h R Iff dom( ) h R dom( ) and ran( ) h R ran( ); where [4], for instance; If ( , , , , , , ( , , then ( , , ) will be added to A BR (Alignment Binary Relation Matrix). 5. Finally, the concepts not yet aligned, will be matched via the terminological methods.
We have adapted the format of the alignment result to the reference alignments restricted to name classes, using the "=" sign for equivalence relation with confidence of 1. Although our system provides other relations as subsumption, and binary relations without measure, as well as the alignment of binary relation by the HR.
In this version we wish to test the techniques used by ONTMAT, such as, the inferences mechanisms applied onWordNet and the ontologies source and target, and the deduction of the matching among entities based on instances. The most appropriate track to do these tests is the conference track. Conference track comprises 16 ontologies from the domain of conference organization. Most ontologies of this track were equipped with OWL DL axioms; which is useful to test our inferences approach.
To improve our application, we will also align the properties of ontologies ( , ). Then, adapt it to read all files type, and integrate the translator to test our tool under other tracks as: Instance Matching, MultiFarm.
The application seals-omt-client from seal, only test files where the alignment relation between concepts is itself the equivalence relation. However ONTOMAT, offers other possibilities in terms alignment relations between entities such as; & : Fuzzy and binary relations. We hope that OAEI takes into consideration those types of relations in the reference alignment file.
We have briefly described the mechanisms exploited by our proposition ONTMAT, and presented the results obtained under the conference track of OAEI 2017. This is our first participation in OAEI, the results are not satisfying, and the system presents some limitations. In the future, we will make great efforts to improve ONTMAT results, and participate in more tracks.