Mapping & Merging Of Heterogeneous Ontologies Based On Fuzzy Similarity And Verificaiton

Ontology is a clear description about the concepts and their relationship within a specified domain.It also serves as a basis for inter-operation among different information systems.These systems are normally developed by different organizations in different times under different requirements,it is, therefore,heterogeneities are inevitable among different underlying ontology. These heterogeneities bring obstacles for required inter-operation among different systems.To solve this problem,the heterogeneities and conflicts among the different ontology need to be identified.They are then used to construct mapping relationship between the corresponding concepts and relations of different ontology(ontology mapping),and to perform merging of different ontology(ontology merging).To improve the efficiency and accuracy of manual ontology-mapping and merging,and to meet the dynamic,real-time inter-operation requirements of different information systems in today's fast changing environments,many research activities on automatic(or semi-automatic) ontology mapping & merging have being carried out both overseas and domestic.Although there have been many progresses and break through,the overall efficiency and accuracy still need to be greatly improved.This dissertation focuses on the improvement of those aspects around ontology mapping and merging. Another important aspect of this research is about the formalized verification method about their logical validity of those results.Major creative thinking and innovative work are summarized as follows:(1) A New Fuzzy Similarity Description MethodThe feature information of an ontology-concept could be classified as text feature information(names & instances etc) and structure feature information (object attributes,taxonomies and dependence functions etc).It is a common practice that,during the process of concept mapping,we need to evaluate the similarities of two concepts for different text features as a set of intermediate solutions.They are then used to composite a suggested solution for the similarity of the two concepts.Followed by sequential iterative processes in which structure features are used to improve the accuracy of the similarity mapping.In every step of this process,all the results(intermediate and final results) are represented by a definite number which is used to compare with some predefined thresholds to determine whether the two concepts can be mapped to each other.Due to their intrinsic fuzzy property of the similarity between two concepts(the so-called fuzzy property is that the similarity between two concepts falls into a certain range in a way described by a probability density function),if we replace these probability density functions with definite numbers too early in the process of mapping compuataion,many useful information will be lost in the later decision making process and the risk will be enlarged.In addition,for these two kinds of ontology features,many traditional approaches adopt the serial method ("Firstly carrying out the mapping based on text information,secondly finishing the iterative verificaiton based on structural information") to perform the computation,the result of the similarity inference based on text information becomes the transcendental knowledge of the iteration based on structural information.Therefore,the effect of the text information may be excessively emphasized,and the structural information contribution to the similarity computation may be weakened.In order to improve the mapping effect,fuzzy similarity method is imported to denote different similarities based on diversified features.Then,the fuzzy property could be used in the process of similarity computation.It avoids the risk of making judgment to fuzzy property too early,and text information also could be concurrently analyzed together with the structural information.(2) A Mapping Method Based on Joint Distribution of Attribute ValuesConcept instance contains a lot of useful information about ontology mapping.In a traditional instance-based approach,the values of each attribute of an instance are directly concatenated into a long text.This approach simply assumes that all of those attributes are independent of each other and without any consideration that there might be some corresponding probabilistic dependent relationships.It is,therefore,some of the important instance-based information is ignored,and the accuracy of the mapping result is debased.To improve the mapping results,a new approach based on the joint distribution of attribute values was proposed so that those dependence relationships could be utilized.In addition,due to the changing environment of dynamic applications, the feature contents of an ontology concept may also change occasionally (such as:add or delete an attribute).It creates incompatibility between the attribute set of primary instances and the changed concept expression,and will create a problem called instance attribute value imperfection in the process of related ontology concept matching.Consequently,mapping result of this joint distribution of attribute values is less trustable.In order to improve its mapping effect,a rough set theory is used here to imitate the sample instance spaces with attribute value imperfection.(3) A Novel Formalized Verification Method for Ontology Merging and MappingDuring a complicated ontology merging and mapping process,many results about new merged ontology and matched concepts will be obtained.It is necessary to check the logical validity of those results.Up to now,most of formalized verifications about ontology are focused on the single concept or relation only.In reality,however,there are many connections existed among different concepts and relations which could be used to facilitate the verification process.Simultaneously,the current ontology mapping result verification is still short of system and effective formalized approach.In this research,an Object Constraint Language(OCL) is adopted from Object Oriented technology and extended to describe the formalized logical relationship among concepts and relations(such as:the structural & constraint of the ontology model and the mapping constraint rules).Based on these formalized logical description,an OCL-based ontology definition meta-model and mapping meta-model have been defined to facilitate related formalized verifications.