Ontologies as the main infrastructure to represent data in the Semantic Web are widely developed independently in each area. When it comes to nding a suitable ontology for a given application, two problems occur: Often, an ontology will cover just a part of the domain of interest or competing ontologies modeling the domain from di erent viewpoints exist. Thus, before being able to leverage the power of ontologies, they themselves need to be integrated. This is a challenging task. The existing approaches are mostly limited to a binary merge. However, by the large availability of the relevant ontologies in the desired domain, an e cient multiple ontologies merging technique is often a necessity to overcome the scalability problem. This research thus advocates for the development of a holistic, e cient multiple ontologies merging method called CoMerger, to satisfy the scalability issue. For e cient processing, rather than merging a large number of ontologies, we merge a small number of clusters. To approve the feasibility of our approach, we will run CoMerger on real-life datasets. Further, our platform will be freely accessible through a live portal.
Ontologies are the semantic model to represent data on the Semantic Web. Often a domain has more than one "standard" ontology for the same general concepts. They either cover just a part of the domain of interest or model the domain from di erent viewpoints. In this fashion, multiple heterogeneous ontologies are independently developed in each domain. In real-world applications of the Semantic Web, this is an essential demand to interoperate with more than two ontologies toward acquiring the desired knowledge for scientists. Indeed, di erent ontologies cover particular aspects of a domain of discourse but overlap to a certain degree. Therefore, an e cient technique for merging multiple ontologies is often a necessity both during ontology development and when ontologies are used in conjunction with data at the query processing level. This can be a cost-e cient approach and saves a lot of development e ort. Thus, before being able to leverage the power of ontologies, they themselves need to be integrated. This is a challenging task.
Existing ontology merging approaches [
Research Question: How can the holistic approach be applied for merging n ontologies to overcome the scalability problem?
To contribute to the main research question, two RQs with further sub-questions can be concluded.
RQ1- How can the elements of input ontologies be e ectively clustered into k clusters based on the detected correspondences to facilitate the merge process? RQ1-1- Does the clustering process lead to a signi cant reduction in execution time and complexity of merging process without compromising quality? RQ1-2- What is the e ect of varying the number of clusters k on the merge quality? RQ2- How can the combination of k clusters e ciently generate the merged ontology?
RQ2-1- Which requirements should be considered for the merging step? RQ2-2- How to ful ll these requirements? RQ2-3- How can a high quality of the merge result be achieved? RQ2-4- How to accomplish consistency in the merged ontology? 2
Merging strategies basically have been divided into two main categories [
To process multiple ontologies, for instance, in the multiple ontologies
matching scenario in [
Principally, general processes in the existing approaches seem to indicate
two di erent strategies: "one-level merge" and "two-levels merge". In the latter
one, an intermediate merge result is produced at the rst level. Then, in the
second phase, the intermediate result is re ned to produce a nal merge result. In
contrast, one-level merge tends to produce the merge result in one incrementally
processing step [
This research aims to comprehensively address the holistic multiple ontologies merging issue to improve the shortcomings of previous approaches. We have been investigating the existing methods and found the scalability issue as an ongoing challenge. Therefore, we have four main objectives: { Overcoming the scalability problem in merging multiple ontologies: We aim to develop a framework for holistic multiple ontologies merging to overcome the scalability issue. { Achieving a high accuracy in the merge result: In order to gain a high accuracy, the merge requirements should be ful lled, however, it should be possible to customize them to the task at hand. We plan to derive a method to apply these customized requirements in a way that they do not contradict each other. { Developing a web tool: As a proof of concept, this research aims to develop a web tool for merging multiple ontologies. This tool can be divided into two sections: merger and evaluator. The latter includes systematic criteria for evaluating the merged ontology independent from the merge techniques. { Validating: To evaluate our framework, we will carry out a set of experimental tests to analyze the performance of the tool and the merge algorithm. 4
Ontology merging is the process of building a new coherent ontology (called
a merged ontology) from given input ontologies to provide a uni ed access
on the domain. Principally, it requires to know which elements are equal
to each other. This can be achieved by an ontology alignment method to
detect the corresponding pairs. Many signi cant advances such as [
Below, a schematic of our holistic multiple ontologies merging framework is illustrated in Fig. 1. It gets as an input a set of ontologies with their alignments, and automatically generates a merged ontology with a set of output mapping between the merged and the input ontologies. In the preprocessing step, the input ontologies and the pre-determined mapping are imported into a repository. Afterwards, the elements of the n input ontologies will be clustered into the k clusters in the clustering step. Finally, the combination phase will be applied to combine the k created clusters into the merged ontology. Thus, instead of the quadratic merge process, this holistic approach needs k merge operations. Here the number of clusters is noticeably smaller than the number of input ontologies (k n). Therefore, to overcome the scalability problem in this framework, rather than merging a large number of ontologies (n), we merge a small number of clusters (k). Once the merged ontology is created, the output
mapping between the merge and the input ontologies will be produced through a backward process in the postprocessing phase.
Contributing to RQ1, the clustering component in Fig. 1 is accelerating the combination process by minimizing the search space in the way of breaking merging n ontologies into the merging of k clusters. Generally, clustering z elements of all input ontologies into k clusters needs z(z2 1) times comparisons. Alternatively, would be comparing z k elements with only k elements, which it requires z k times comparisons. Therefore, we will use a core-clustering method, where all z elements will be compared with k core of the clusters through a semantic similarity function. We deem that the elements with a high number of correspondences in the input ontologies are more suitable to be considered as the cores. In this follow, the core of each cluster and the optimal number of clusters for each set of ontologies will be dynamically determined by using their mapping information.
To address RQ2, we divide the combination process into two steps through the combining component (Fig. 1): (i) intra-combination, and (ii) inter-combination. In the rst step, the elements inside the clusters will be combined to create k sub-ontologies. In the inter-combination step, the k generated sub-ontologies will be attached to create the nal merged ontology. Here, detecting the nest place of the join is a challenging task. We intend to nd it in a heuristic method by narrowing comparison between the leaves and the core of sub-ontologies.
To capture the accuracy and consistency of the result, the
predetermined merge requirements will be assessed before applying each
combining process. The existing merge technique partially aims to satisfy some
of them, however, they should be customized by the task at the hand (RQ2-1 ).
To this end, we are providing a checklist including a variety of requirements
(extracted from [
To develop a web-based tool, a preliminary version of CoMerger is being provided by using the OWL API library 2 and a user-friendly GUI as is shown in Figs. 2 and 3. The further version might also include the analysis the execution log of the merge evaluator and visualization of the results. Evaluation Protocol: We address the required evaluations as below: E1. To study our main research question, we will compare the quality and complexity of merging multiple ontologies by a series of the binary merges rather than our holistic merge approach. The quality will be measured as states in E5, and the complexity will be measured as the number of required operations.
E2. The aim of clustering is to accelerate the merge process. Therefore, to
evaluate RQ1-1, we will compare the quality of the merge result and
complexity of the merge operations with and without using clustering.
E3. To evaluate our heuristic approach on RQ1-2, we will compare the quality
of the merge result on the experimental tests for k = 1; :::; n. Besides, the
cohesion and coupling [
E5. The quality of the merge result regarding RQ2-3 can be evaluated in three
scenarios: (i) Measuring the integrity of the merged ontology with the merge
quality criteria, namely, compactness, completeness, and minimality [
E6. The consistency of the merge result can be evaluated by measuring the ful llment of the customized merge requirements with either a reasoner or an expert (RQ2-4 ).
E7. We demonstrate the method's scalability by illustrating the performance test results on the set of real-life ontologies from BioPortal repository 3 and OAEI datasets. The rst one currently contains more than 700 biomedical ontologies with thousand of classes, and the second one includes several domains such as biodiversity and ecology, anatomy and conference domains. Here, the runtime performance will be evaluated based on the number of ontologies versus the time required for merge operations.
The rst conceptual ontology-based data integration work ow has already been
represented in [
Ontology merging is often a necessity in applications of the Semantic Web. To
this end, our aim is to provide a holistic multiple ontologies merging, namely
CoMerger, to satisfy the scalability issue. The e cient processing will be held
by breaking the n ontologies processing into the k clusters merging, with a
minor overhead of the clustering process. Each component has a high e ect on
the quality of the nal merge result, therefore, the di culty of this research
would be carefully ne-tune the correctness of each sub-function. Besides, in the
aspect of knowledge integration, the possibility of ambiguous knowledge being
introduced in the nal merged ontology will be another obstacle in this research,
which bring us to deal with the reasoning from ambiguous knowledge challenge
[
The author would like to thank Prof. Birgitta Konig-Ries and Dr. Alsayed Algergawy for their valuable supervising. The author is supported by a scholarship from German Academic Exchange Service (DAAD).