Research Of Ontology Matching Algorithm

This paper is supported by National High Technology Research andDevelopment Program of China (Research and Application of Digital AgricultureKnowledge Grid, 2006AA10Z245). Our research area is how to create domainontologies, and how to calculating the similarity between them. Our aim is to achievethe desired goal of the matching, and then we can use the existing network resourceswith the maxmum.Nowadays, the computer is changing from a singal device to the worldwildnetwork which has ability to exchange information and deal with transaction,therefore, the reusing and sharing of the data and information is the urgent task for thecomputer technology. So ontology research and development has become aninevitable trend in the field of knowledge engineering, natural language processing,information retrieval systems, intelligent information integration, knowledgemanagement, information exchange, software engineering and so on.Ontology in the Semantic Web, E-commerce and other areas has a much wideapplications. The ontology's greatest contribution is that it can make the concepts andterms in one or many specific areas standardization, and it can facilitate the practicalapplications in and between the fields. However, different organizations or individualsbuild their ontologies without the uniform standardization, this has led difficulty tointegtate and share between ontologies, so ontologis in the same field matching andmapping is becoming a research direction.The main body of this article is about the ontology construction and ontologymatching algorithm. Firstly we introduce some basic knowledge about ontology,including the background of ontology, the history of the ontology development, andthe current research status. Secondly we elicit the definition of ontology, ontologydescription language (such as RDF, OIL, OWL, etc), construction rules of ontologyand how to build the ontology with the software Protégé. Thridly we recall some basicontology matching techniques, including String-based techniques, Language-basedtechniques, Constraint-based techniques, Linguistic-based techniques, Alignmentreuse techniques, Graph-based techniques, Taxonomy-based techniques, Repository of structures, Model-based techniques and so on. Fourly we sum up some existence ofimportant ontology system (e.g. WordNet, HowNet, etc). The finally we introduce anumber of classic matching systems (e.g. Cupid, COMA, S-Match, etc) and discusstheir used algorithms, the implementation process and the specific matching process.Learning through the above-mentioned, we can put the ontology as a graph withsemantic information, the edges and nodes of the graph describe its semanticinformation. So we can deal with the ontology matching as the graph matching thathave the semantic information. Therefor, the matching between concepts fromdifferent ontologies can be discovered by their semantic similarity, and also inexisting approaches, similarity often derives from instances, relations and informationof hierarchy of concepts. After research we have discovered that the above threereferenced objects reflect the relationship among concepts in some degree.Based on the above idea, the paper have researched the ontology matchingalgorithm which is based on the semantic, we integrate the knowledge of statistics andgraphics. First of all, we need to define some terms and formulas that we used in thispaper, and then we introduce a new ontotogy matching algorithm based on semanticwhich is called OMABOS. We set up original model of the OMABOS and comparethe results between falcon-AO and OMABOS. In this approach the similarity ofconcept can be defined by joint probability distribution of the concepts and instances.In the same way, the similarity of relation can be gotten by statistics and graph theory.The similarity of next concept can be found under the guide of the relation matchingrule, and then the similarity between concepts can be obtained by calling the functionof concept similarity recursively. Finally all the values with weights of the conceptsimilarirty of every hierarchy should be added and then it can be integrated into thesemantic similarity of two ontologies.OMABOS contains the main algorithms are as follows:1. Ontology Preprocessing: we adopt the Jena package to do the ontologypreprocessing. We use the graph to express the ontology, the nodes describethe concept or the instance, and the edges describe the attribute.2. Font similarity calculation: the ontology's concept will be decomposed into acollection of the basic terms, and we adopte the method of Edit distance tocomputer the font similarity of the basic terms in the two collections. Herewe use the formula of the falcon.3. Meaning similarity calculation: we adopt the WordNet as the lexicon. We use the synonyms that are defined in it, and consider the depth of the word.4. Element-level similarity: we combine the font similarity and the meaningsimilarity which are mentioned above through the definition of the formulaand the correlation coefficient which is obtained from the field experts, thuswe can calculate the number of the shared words.5. Structure similarity calculation: This is an algorithm which is proposed bythis paper. First of all, we adopt the root node of the ontology as the entrancenode, calculate the number of the shared words and the depth coefficientbetween them, and then return the level similarity through them; here thedepth coefficient is based on the depth of the concept in the ontology.Secondly, on the basis of the two nodes, we calculate the similarity of theedgs which are starting from the two nodes, if the similarity of the edge isnot 0, then begin the process of the recursion, the nodes which are the edgespointed to is as the next entrance nodes. The finally, we add up all thenodes'similarity and edges'similarity with a weight, and then get the finalsimilarity between the two ontotogy.6. We can put every concept in this ontology as a root node of sub-ontology,and use the above method of calculation to acquire the semantic similaritybetween each pair of the concept, for leaf nodes, because there is no longeredgs from it, we can simplify the calculation. The finally we output everypair of similarity with the triple approach.We build two simple domain ontologies with the software Protégé, and doexperiment to verify our approach. We firstly use the falcon-AO system which isinvented by Southeast University to calculate the similarity of the two ontologies, andrecord the experimental results which are expressed by the pairs of the matchingconcepts and their similarity. Then we use our method OMABOS, also record thepairs of the matching concepts and the similarity between them, the finally we analysethe results of the OMABOS, enumerate a pair of concepts which have similarity intheory, but the falcon-AO system can not get this result, and give the steps of theprogram, from this we can prove the correctness and implementation of our algorithm,we say our method OMABOS can get more pairs of the matching concepts in thecase of accuracy.