This paper presents an FCA-based methodology for concept detection in a flat ontology. We apply this approach to an automatically generated ontology for a RosettaNet Partner Interface Process (PIP) which does not take advantage of some important OWL semantic relations like subClassOf. The goal of our approach is to regroup ontology classes sharing a set of properties (Data and Object Property) in order to improve the quality, readability and the inheritance richness of a flat ontology.
Ontologies are widely used in knowledge management, information integration,
natural language processing, information retrieval, business-to-business (B2B),
e-commerce, etc [
Research is now envisioning the adoption of semantic web technologies in the
business domain. Lytras et al. [
After an ontology is constructed, it is important to assess the quality of
an ontology to detect design defects and to automatically recognize parts that
cause problems and might need more work. In this paper, we try to improve the
inheritance richness of an OWL ontology based on the balance between depth
and height of the inheritance tree which, according to Tatir et al. [
In this paper, we propose a method to verify, regroup concepts sharing properties to improve the quality and readability of an automatically generated ontology. We use Formal Concept Analysis in the context of concept detection in OWL ontologies using OWL artifacts (Class, DataProperty and ObjectProperty ).
The remainder of this paper is structured as follows. Section 2 states the problem. In Section 3, we describe the principles of the Formal Concept Analysis algorithm underlying our approach and the adaptation of its principles for the detection of concepts in a RosettaNet Ontology. Section 4 presents and discusses the validation results. A summary of the related work in ontology-based attempts of B2B standards and the use of Formal Concept Analysis in Ontological Engineering is given in Section 5. We conclude and suggest future research directions in Section 6. 2
In this section, we describe the problem of concepts detection and the importance of the inheritance tree to represent a domain knowledge. We start by defining important notions. Then, we detail the specific problems that are addressed by our approach. 2.1
Ontology in Semantic Web technology provides a shared and common
vocabulary for a domain of interest and a specification of the meaning of its terms [
RosettaNet B2B Standard is a consortium which provides a global forum for suppliers, customers, and competitors to do business and collaboration in an efficient and profitable manner. To manage business activities, RosettaNet formalizes Partner Interface Processes (PIP) with either Data Type Definition (DTD) format or XML Schema, and define business processes between trading partners. PIPs are organized into eight groups of core business processes called clusters, themselves further grouped into segments. Each segment includes several PIPs. In section 3, we will use the PIP3A4 as running example of our methodology. But we managed to apply the same technique on all published PIPs.
The RosettaNet architecture contains a Business Dictionary to define the
properties for basic business activities and Technical Dictionaries to provide
properties for products [
There are few works to measure quality of an ontology using metrics. But, to our knowledge, no work has yet proposed methods to verify, maintain concept hierarchy representation and improve the quality and readability of an ontology from inheritance point of view using OWL artifacts (Class, DataProperty and ObjectProperty ).
Tatir et al. [
In our case study, we have chosen an ontology describing a detailed domain knowledge. Some approaches are proposed to ontologize some of the famous B2B standards like RosettaNet1 and ebXML2.
In this paper, we apply our methodology on a RosettaNet Ontology described
in our paper published in the 15th International Conference on Enterprise
Information Systems [
During the transformation process, it is difficult to automatically detect elements with a common semantics which are described in separate OWL files. But, we noticed that many of those share some properties and lexemes in their compounded names. For this reason, we use FCA as a classification approach to detect concepts by regrouping classes in an ontology to improve its quality and readability. 3
The goal of our approach is to regroup ontology classes sharing a set of properties (Data and Object Property) and then maintain the hierarchy representations.
The core of our system has three main parts: Ontology Processing, FCA System and Regrouping concepts. The Ontology Processing step builds a cross-table from ontology artifacts (Class, Data Property and Object Property) without dealing with the property type or occurrence restrictions. This table describes relationships between objects (ontology classes are represented by rows) and attributes (Data and Object properties correspond to columns).
Taking the example in Table 1, o1::k represents ontology classes with k is the number of classes in the ontology. p1::m represents the set of Data and Object
Properties with m being the total number of data and object properties. Element
<i,j > of the table is marked with “ x” if the domain of property pj is class oi.
We use FCA to create a hierarchy of concepts displayed as an inheritance tree.
According to [
The concept lattice in Figure 1, which is the output of our FCA application, describes the RosettaNet PIP3A4 (Purchase Order Request). Intermediate concepts (of the form Concept_NN) were generated by the algorithm and represent the shared set of properties. The others represent the original ontology classes.
Each class will be associated with the concept having the largest number of shared properties. As shown in Figure 1, we note that the Concept_7 combines the two classes ProductSubLineItemType and ProductLineItemType with 19 shared properties; Concept_8 joins ServiceLineItemByOptionType and ServiceLineItemType because of 22 shared properties.
The combined classes share some semantics consistent with the concept definitions provided by RosettaNet. They represent one concept (Concept_7 and Concept_8) that has been generated by our approach. As the regrouped classes share tokens in their compound names, we will use these to rename the FCA generated names. So, Concept_7 becomes ProductLineItem and Concept_8, ServiceLineItem.
Using the generated group of classes, we update the original ontology. To do this, we used the OWL API3 for manipulating, developing and maintaining the ontology. 4
The goal of our study is to evaluate the efficiency of our approach for concept detection in an OWL ontology. In this section, we describe our experimental setup and results.
In order to test our methodology, we use the RosettaNet Ontology, proposed
in our ICEIS 2013 paper [
In Table 3, we evaluate our RosettaNet Ontology using state of the art
ontology metrics defined in Table 2. Only basic metrics related to the main elements
of ontologies have been used [
Metrics Definition classes (noc) number of classes (|C|) in the ontology. data properties (nodp) number of data properties. foobject properties (noop) number of object properties. rebmspurobpclearstsieess ((nnoopsc)) snuummboefrnoofdspuabncdlasnsoeosp(|msCe|t)riicns.the ontology. uNroot classes (norc) onfumthbisermoeftrroicotiscblaestswesee(nwi1thaonudt |sCu|p. erclasses). The range leaf classes (nolc) cbleatswseesenw1itahnodut|Csu|.bclasses. The range of this metric is
We notice in Table 3 that the values of ir and nosc metrics are zero and the values of norc and nolc are equal to the number of classes |C| in the ontology. These metric values indicate that our original RosettaNet PIP Ontology as generated from the DTD and XML Schemas is a flat or horizontal ontology representing a general knowledge despite the fact that RosettaNet represents a specific domain of interest with many elements sharing a common semantics.
The application of our FCA-based Concept Detection approach extracts 182 groups of concepts comprising 384 classes from the 1252 in the ontology, bringing the inheritance metric ir from 0 to 0.31. On average, each concept contains 2 classes.
We extracted 90 groups of concepts because several concepts are shared between PIP files e.g. the concept, combining RegionalBusinessTaxIdentifier and NationalBusinessTaxIdentifier classes, is detected in 3 PIPs (PIP3A5, PIP3A11 and PIP3B6).
We also performed a manual validation of all detected concepts. We noticed that among the 90 concepts detected, 79 concepts have effectively a common semantics. So, we have a detection precision of 87%. From Table 3, we can see that the value of nosc and ir metrics increase, norc and nolc decrease after applying our FCA-based approach.
Ontologies and Formal Concept Analysis (FCA) aim at modeling concepts [
Some FCA-based proposals in Ontology Engineering differ from our own
strategy with respect to the nature of the problem. Cimiano et al. [
Stumme [
Obitko et al. [
We have presented an approach for concepts detection in a flat ontology using a Formal Concept Analysis and applied this methodology on a RosettaNet PIP Ontology which was automatically generated ontology.
Our goal is to improve the readability, maintainability and inheritance richness of an ontology. We used FCA to detect groups of concepts sharing properties and having a common semantics. Through the use of ontology metrics, we have shown that our FCA-based concept detection methodology can improve inheritance richness.
As the results using RosettaNet are promising, we suggest as future work to test the efficiency of our concept detection approach to other ontologies dealing with other domains.
We would like to thank RosettaNet Group for allowing us to download the RosettaNet Partner Interface Processes (PIPs) from the RosettaNet website. This work has been partially funded by Tunisian Government and NSERC.