Grid Computing for Ontology Matching

Introduction

Nowadays, lots of different ontologies as well as tools and frameworks that support matching are available. However, the amount of available semantic data structures is significantly growing. This issue faces the challenge of ontology matching in a scalable way considering large scale ontologies. For this, a strategy for handling big data sizes in an efficient way is required. Further, the field of bioinformatics is a beneficial use case because of the amount of available domain ontologies such as from the official NCBO BioPortal website (The National Center for Biomedical Ontology, 2009) and the need to examine several ontologies to solve a specific question.

The complexity of matching large-scale ontologies and ontology matching in urgent computing use cases entails the problem of matching in a scalable way. Hence, distribution techniques such as grid computing are used to increase scalability by executing it by the use of distributed heterogeneous computing resources. Hence, the aim of this work is to support ontology matching strategies with a grid architecture to provide required computing resources for compute resource intensive and urgent computing use cases. The ontologies that are considered for this work are mostly OWL ontologies.

Furthermore, the LarKC project (The LarKC project, 2010) in which new techniques for processing large datasets are developed for the usage of concrete use cases such as “Semantic Integration for Early Clinical Development” or “Carcinogenesis Reference Production” is a beneficial basis for this work. Within LarKC project ontologies are used as well and new techniques usage of large-scale data sets are developed and used for real time applications such as an urban city use case that requires continuously semantic data with the aim to analyze such data in a time-saving way. For use cases as they are mentioned above, sophisticated techniques are required to run several processes at same time in a distributed fashion as it is done in the grid. At this, the new idea is to set one ontology from a given set as the priority ontology which is enhanced by matching the concepts of the priority ontology with the concepts of other selected ontologies from a given set.

This work supports an adequate matching procedure as well as a mapping of similar concepts or properties of the ontologies in a grid. For this, the first step is to define an adequate architecture which meets the requirements of ontology matching in a grid. After the matching is executed the ontology parts and the matching results are merged together in the priority ontology with the aim to extend this ontology. However, the matching strategy will be aligned to the grid architecture. The presented idea for matching ontologies in a grid environment is an effective method to solve the challenge of matching in a scalable, robust and time-saving way. Further, it is of interest to examine the principles and design issues for this topic.

Currently, lots of ontology matching strategies have been published so far. Hence, a clear distinction between this work and current approaches is required. This work aims to map concepts, properties as well as relations between entities of several ontologies in a semi-automatic manner by considering well known developments in this field and extension of these approaches. Further, the ontology matching is supported by a distributed architecture in order to use several compute resources. Beyond the distribution of ontology matching processes, ontologies are selected with the aim to enhance one ontology to a priority ontology of a given set. In order to achieve an adequate mapping of similarities for the enhanced priority ontology similarity values for entity pairs within the matched ontologies are calculated. However, the calculation of similarity values for entity pairs encounters the problem of evaluating one measurement out of lots of values such as similarity of properties or relations and relevance to neighboring entity pairs and much more.

Further, as mentioned the matching of large data sets requires a high amount of compute resources as well. For this, processing algorithms at same time on several nodes (e.g. cluster architecture) is a beneficial solution. However, this raises the challenge of distributing the jobs in a high effective way as well as aligning algorithms to avoid latencies or conflicts. Therefore, two main issues are considered in this work: (i) Ontology matching based on similarity values, and (ii) Distributed ontology matching on several resources by usage of a grid architecture.