In this paper we present a tool, SOMET, for collaborative developing, matching and merging ontologies. The tool's design is based on a Wiki model, allowing for multiple authors to contribute to an ontology. It also provides a number of meta-ontology features, including the ability to compare, match and merge. The tool makes use of one algorithmic approach to element-level mapping, demonstrating the use of both automated and manual matching.
http://www.rmit.edu.au
Much research has been invested into automated techniques of ontology matching
[
This paper presents an online tool, SOMET (pronounced ‘Summit’), for collaborative developing, matching and merging ontologies. It is motivated by the idea that ontology development and matching is essentially a social and interactive activity, and is best served by tools which permit this. As such, the tool is based on the Wiki model for ontology development.
This paper is structured as follows: In Section 2, we review related work.
Section 3 presents a case for collaborative ontology development and matching.
Section 4 describes the design and implementation of the SOMET prototype,
along with our approach to ontology matching and merging. Section 5 examines
the test results of using SOMET on two sample ontologies. In Section 6, we
look at further directions for SOMET. Section 7 concludes the paper.
The concept of shared or collaborative ontology development is not new. A
number of tools have been introduced, including CODE [
There has been considerable exploration of approaches for ontology
matching. Much of the research has been into finding suitable algorithms to automate
part or all of a given matching task. As shown by surveys [
Recent work also suggests the importance of community-driven ontology
matching [
Ontologies are typically developed by a process of iterative construction and consultation, with a focus on concepts in a specified domain. For the most part construction and consultation are separate activities, conducted in serial fashion, as the modeller uses specific knowledge of the modelling environment to apply the results of the consultation process. In the case of traditional database and software engineering activities, such established practices are generally entrenched in broader lifecycle processes, and there has been little impetus to shift the onus of model construction from the modelling expert. For Semantic Web ontologies, where models are frequently shared across organisational and geographical boundaries, there is significantly greater motivation to develop such models collaboratively.
The approach used in this paper is based on the success of open access content systems. The approach accepts that a lower grade user interface will be acceptable in certain contexts, just as authoring content online is an acceptable degradation from using a word processor in certain contexts. In particular, if the ontology is small, then the frequent submission and retrieval of ontology elements across a network of ontology components will be tolerable. 4
SOMET is a prototype that has been developed in Ruby, using the Ruby on Rails framework. It employs a commonly used model-view-controller architecture. It also employs a plug-in architecture for executing matching algorithms.
The SOMET interface makes it possible to construct an ontology with classes, properties and individuals. Notable features at this stage include the following: – Creation and editing of ontologies, classes, properties, individuals and annotations. – Importing and exporting an ontology. – Generation of a comparative report of differences between two ontologies. – Manual and semi-automated matching of classes. – Merging of two ontologies. – Various Wiki features, such as publishing, sharing, publishing, versioning, logging and commenting on ontologies. Class and property matches can be proposed, discussed, and approved or rejected.
We have conducted a partial implementation of the S-MATCH algorithm
[
Step 3. For all pairs of labels in the two trees, compute relations among concepts of labels.
Step 4. For all pairs of nodes in the two trees, compute relations among concepts of nodes.
For our purposes, we utilised WordNet in steps 1 (to develop the senses of each lemma) and 3 (to generate the relations between pairs of labels). The generation of the label matrix at step 3 made use of an exhaustive traversal through the WordNet dictionary. The results of step 3 are a set of tuples, heID, n1 i, n2 j, Ri, where eID is a unique identifier for the given element-level match, n1i is the i-th node of the first ontology graph, n2j is the j-th node of the second ontology graph, and R is one of equivalence, more general, less general or disjoint. We ignore possible overlapping or unknown relations. These tuples are captured in a database, grouped together as an ontology match.
The Merge operation generates a new ontology graph from two existing ones, on the basis of a defined ontology match. The resulting merged ontology is stored in the database. The operation has 5 steps: 1. Compute the set of direct child-parent relations for the new graph, based on the set of element-level matches. 2. Compute the set of direct parent-child relations for the new graph, based on the results of step 1. 3. Generate the complete set of parent and child nodes for the new graph, based both on relations from steps 1 and 2, as well as the existing relations in the source graphs. 4. Compute the set of disjoint relations for the new graph, based on the set of element-level matches, for all siblings in the graph generated by step 3, where such siblings are not already disjoint. 5. Perform a deep copy of non-matched objects from each of the source ontologies into the new ontology. 5
The matching and merging capabilities of SOMET have been tested using two simple ontologies. The goals of the test were to successfully invoke the S-MATCH implementation, generate a set of class-level matches, add or modify one such match, and to merge the two ontologies into a third on the basis of the matches. The test would be successful a) if the merged ontology contained the union of the two source ontology classes in a directed acyclic graph, with at least some successful matches, and b) if the Match and Merge operations execute in reasonable time. Tests were conducted on a P4 3.0GHz machine with 2GB of memory.
The matching operation took 74.407 seconds. Further analysis showed the majority of this time was due to the exhaustive scansion of the WordNet database. The merging operation took 5.703 seconds. The results show some inconsistencies of the WordNet associations, with certain anomalous subsumtion relations identified.
The results of the test show that for small ontologies, the Match and Merge operations can be conducted in an shared online environment. While the performance is sub-optimal, this could be corrected by depth-limited WordNet searches, and augmented with domain-specific vocabularies. The quality of the match also varies, although implementing step 4 of the S-MATCH algorithm, ie. the satisfiability checks on formulas representing the full path of any given element, would improve this greatly. The test also demonstrates the ability to augment automated matching techniques with manual matching. 6
SOMET has been developed to a prototype level, and as such lacks the kinds of user interface enhancements expected of such a tool in a production environment. To realise the aims of providing a generic matching tool with ‘pluggable’ matching techniques, a more sophisticated plug-in architecture needs to be developed. Given the diversity of matching approaches, the viability of this aim also needs to be better ascertained. Research on publishing and document standards over several years has motivated the investigation into how ontologies can be developed and matched in a collaborative way. We have concluded social interaction, negotiation and collaboration are necessary aspects to successful ontology matching in many environments. So far, tools for ontology matching have focussed on private ontology matching. In this paper we have presented SOMET as a prototype for collaborative ontology development, matching and merging. The test results have been encouraging in terms of its utility for these purposes.
Acknowledgments. This work has been supported by ARC Grant LP0667834, ‘Towards the Semantic Web: Standards and Interoperability across Document Management and Publishing Supply Chains’.