Pattern-based Ontology Construction

                                  Eva Blomqvist

                     Jönköping University, Jönköping, Sweden
                                  blev@jth.hj.se


      Abstract. Large and complex enterprise systems face the same kind of
      information processing problems that exist on the web in general, and
      constructing an ontology is a crucial part of many solutions. Construction
      of enterprise ontologies needs to be at least semi-automatic in order to
      reduce the effort required, and another important issue is to introduce
      further knowledge reuse in the process. In order to realise these ideas
      the proposed research focuses on semi-automatic ontology construction,
      based on the methodology of case-based reasoning.


1   Introduction
When developing semantic applications for enterprises, constructing the enter-
prise application ontologies is a crucial part. Manual ontology engineering is a
tedious and complex task. Another issue is knowledge reuse, common practises
of the business world should be exploited as well as drawing on best practises in
ontology engineering. By combining patterns with a case-based reasoning view,
we aim at developing a novel semi-automatic ontology construction approach.

2   Background and Related Work
Our research focuses on application ontologies within enterprises, mainly for
structuring and retrieval of information. We view an ontology design pattern as
an ontology template, which is self-contained, comprised of a set of consistent
ontology primitives, and intended to construct a part of some ontology. Related
work on ontology patterns focus mainly on templates for manual use (like in [1]).
    Recent developments in ontology engineering involve ontology learning (OL)
as in [2], [3] and [4]. A major problem is that much of the information in a
company is not explicitly stated, this is one issue where patterns can be of
assistance. Case-based reasoning (CBR) is trying to mimic human behaviour,
using previous experience to solve new problems. A case is a problem situation,
previously solved cases are stored in the case base for reuse. The CBR process
is viewed as a cycle of four phases: retrieval, reuse, revision and retaining cases.

3   Research Hypotheses
In our research some specific research questions have been derived, that can then
be reformulated as the following hypotheses:
 – CBR gives a framework for further automation of the ontology construction
   process, compared to related semi-automatic approaches that exist today.
 – Using the CBR methodology (with patterns) can improve the quality of the
   generated ontologies compared to existing semi-automatic approaches.
 – Automation reduces the total construction effort.
 – Domain knowledge and engineering experience can be reused through pat-
   terns.
   To verify the hypotheses the proposed method must be evaluated and com-
pared to manual approaches as well as the related OL approaches stated earlier.
The result produced by the method must be evaluated and shown to be of better
quality compared to the result of related semi-automatic approaches.

4    Proposed Approach
The basis of a CBR approach is the case base and its content. In our approach
the case base corresponds to a pattern catalogue (pattern base). The design
patterns are represented as small ontologies and the architecture patterns are
sets of constraints on the combination of design patterns, and may also include
connections to specific design patterns.
    The retrieval phase constitutes the process of analysing the input text cor-
pus and deriving its representation, then matching this to the pattern base and
selecting appropriate patterns. The reuse phase concerns the reuse and adapta-
tion of the patterns, combining them into a first ontology. The revision phase
includes extending the ontology, based on evaluation results. Retaining patterns
includes the discovery of new patterns as well as improving existing patterns.
    In our approach there is uncertainty inherent in all the described steps. For
example each ontology primitive of the input representation have a certain degree
of confidence associated, and the patterns are in themselves associated with a
certain level of confidence. The levels of confidence are transferred onto the
constructed initial ontology and can be used when evaluating it.
    The main contributions of this approach is envisioned as both further au-
tomation of the ontology construction process, but in addition an increased
quality of the produced ontology, as compared to other existing OL approaches.
This increased quality will mainly be due to the use of patterns and the presence
of an evaluation and revision phase in the method.

References
1. Gangemi, A.: Ontology Design Patterns for Semantic Web Content. In: Proceedings
   of ISWC 2005. Volume 3729 of LNCS., Springer (2005) 262–276
2. Cimiano, P.: Ontology Learning and Population from Text: Algorithms, Evaluation
   and Applications. Springer Science (2006)
3. Fortuna, B., Grobelnik, M., Mladenic, D.: Semi-automatic Data-driven Ontology
   Construction System. In: Proc. of IS-2006, Ljubljana, Slovenia (2006)
4. Iria, J., Brewster, C., Ciravegna, F., Wilks, Y.: An Incremental Tri-partite Approach
   to Ontology Learning. In: Proc. of LREC2006, Genoa (2006)