Similarity-based Ontology Mapping in Material Science Domain
Wei Lin, Changjun Hu, Yang Li and Xin Cheng
School of Computer and Communication Engineering, University of Science and Technology Beijing,
No.30 Xueyuan Road, Haidian District, Beijing, China
Keywords: Ontology, Mapping, Similarity Measure, Material Scientific Data.
Abstract: How to accurately retrieve data for users from massive, distributed and relational complex material
databases is a major challenge in the domain of material science. Ontology mapping is regarded as a
solution provider to the problem addressed. The number of material ontologies that are publicly available
and accessible increases dramatically, so does the need for establishing semantic mapping among them to
ensure interoperability. In this paper, we proposed a compositive similarity measure for ontology mapping.
The material ontologies are generated from relational databases schemas based on rules. Then they are
compared from concept name, structure and individuals. Finally, we describe a set of experiments on
material science domain and show that our method propose highly accurate ontology mapping.
1 INTRODUCTION
The specialities of scientific data are the barrier of
data sharing. Generally, scientific data has the
following features: 1) Massive: China’s scientific
data size has reached PB scale. 2) Distribution:
different types of data locate in heterogeneous
databases. 3) Relevance: the relevance among
scientific data is complex.
Data retrieval from massive, distributed and
relational complex scientific databases remains an
issue: 1) Traditional keyword search is
unsatisfactory as it only considers the literally
matching of data. 2) Efficiency and precision of data
retrieval is extremely low due to distribution of
material scientific data.
Accordingly, semantic retrieval based on
ontology mapping is recommended. However,
ontology research is inadequacy in scientific domain
due to the mentioned features of scientific data. In
this paper, we propose a compositive similarity
measure for ontology mapping. The material
ontologies are generated from relational databases
schemas at first. Then the similarity of concept
name, structure and individuals between different
ontologies are calculated separately and composited
to a compositive similarity. Finally, we prove in
experiments that the compositive similarity measure
achieves high precision ratio and recall ratio and is
more appropriate for ontology mapping in the
domain of material science.
The remainder of this paper is structured as
follows. Section 2 will provide the reader with
related research while section 3 will explain the
details of our proposed measure. The experimental
data and analysis will be presented in section 4.
Finally, the conclusion of this paper and remarks on
future work are given in section 5.
2 RELATED RESEARCH
At present, there are quite a lot ontology mapping
systems. Doan et al. (2003) classified the individuals
of concepts based on machine learning method and
implemented GLUE ontology mapping system.
Ehrig and Staab (2004) optimized existing ontology
mapping measures and realized a quick ontology
mapping system QOM. OLA ontology mapping
system was developed specifically for OWL
ontologies (Euzenat and Valtchev, 2004). Falcon-
AO ontology mapping system was implemented by
Hu et al. (2005) based on a graph matching
algorithm. These are powerful systems, but suffer
from a few drawbacks. The QOM system only
computes the similarity of a restricted subset of
candidate concept pairs. It’s considered as a way to
tradeoff between quality and efficiency of the
mapping generation algorithms. GLUE system
doesn’t take into account structural information of
265
Lin W., Hu C., Li Y. and Cheng X..
Similarity-based Ontology Mapping in Material Science Domain.
DOI: 10.5220/0004350902650269
In Proceedings of the 9th International Conference on Web Information Systems and Technologies (WEBIST-2013), pages 265-269
ISBN: 978-989-8565-54-9
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
ontology concepts while OLA system and Falcon-
AO system ignore the influence of individual
information. Since results of all these systems are
one-sided, a similarity measure with better
performance is needed.
Concept similarity computing directly relates to
both quality and reliability of the mapping process.
Name similarity computing of ontology concepts is
relevance to word similarity computing. A lot of
scholars have done much work on word similarity
computing based on semantic dictionary. Budanitsky
and Hirst (2006) evaluated five lexical semantic
relatedness measures by comparing their
performance in detecting and correcting real-word
spelling errors. Mahapatra et al. (2010) computed
similarity of two sentences by calculating a weighted
sum of the word similarity and the semantic
similarity. All of the measures use WordNet as their
central resource.
All these above measures are just for word
similarity computing. Concept similarity computing
of ontologies should also consider other affecting
factors such as concept structure and concept
individual.
3 METHODOLOGY
According to the features and requirements of
material scientific data, we constructed ontologies
from relational databases schemas. A compositive
measure is then proposed for ontology mapping
based on their similarity.
3.1 Mapping from Databases
to Ontologies
We define five rules to transform relational
databases to OWL ontology base on the features of
material scientific data and OWL description
language (Cullot et al., 2007).
Rule1: join-tables in the databases convert into
object properties (OWL:ObjectProperty).
Rule2: other tables convert into classes or
subclasses (OWL:Class or OWL:SubClass).
Rule3: referential constraints of tables convert into
object properties (OWL:ObjectProperty).
Rule4: columns of tables convert into data
properties (OWL:DataProperty).
Rule5: rows of tables convert into individuals.
(OWL:Individual).
Figure1 shows the specific rules.
Figure 1: Transformation from databases to OWL
ontology.
Figure 2: Relations of tables.
For example, with table t_grade in the logical
structure shown in figure2, our transformation is as
follows: Convert table t_grade into OWL class
t_grade; Convert foreign key constraint FK_class_id
into object property class_id; Columns of table
t_grade would be converted into data properties and
each row of table t_grade would be converted into
individuals.
3.2 Similarity Computing
Similarity computing is the direct basis for creating
semantic relations between concepts. At present,
similarity computing mainly based on concept name,
definition, structure and so on. Since each measure
has its limitations, the result of each single measure
is one-sided and can not fully reflect the relations
among concepts. For material ontologies, concept
name can reflect information of the concept to a
certain extent. For example, concept name
‘thermoplastic’ means a kind of plastic material.
Likewise, concept structure reflects the hierarchical
WEBIST2013-9thInternationalConferenceonWebInformationSystemsandTechnologies
266
information of ontology. For instance, if a material’s
parent concept is plastic, we can infer that it must be
a branch of plastic. Each individual represents a
specific material in material ontologies. Therefore,
we conclude a compositive measure according to the
above-mention features of material ontologies.
3.2.1 Name Similarity
In traditional similarity measures, name similarity
depends on the matching degree of name string.
However, this measure only considers literally
matching when calculating similarity.
Therefore, similarity computing based on
semantic dictionaries improves accuracy of
similarity. In general, for two given concepts, the
shorter the path in WordNet tree, the higher the
similarity. However, the depth of a concept in the
hierarchy tree is ignored.
According to the features of material ontologies
and word organization rules of WordNet, we
propose a similarity measure based on WordNet.
Both depth and distance of concept pairs are
considered in this measure, the formula is shown as
the following:
Sim
C1, C2
2∗depcomC1,C2
2∗depcom
C1, C2
dis
t
C1, C2
(1)
Note that com(C1,C2) represents the least common
node of C1 and C2, dep(com(C1,C2)) represents the
depth of com(C1,C2) in the hierarchy tree,
dist(C1,C2)represents the shortest distance of C1
and C2 in the hierarchy tree.
3.2.2 Structural Similarity
For material ontologies, concept structure reflects
the hierarchical information of ontology. For two
concepts in material ontologies, if their parent
concept is similar, they are similar to a certain extent.
Likewise, if their son concept or brother concept is
similar, they are possibly similar. The formula is
shown as the following for calculating structural
similarity:
C1,C2
ω
Si
C1, C2 ω
Si
C1, C2 ω
Si
C1,C2
ω
ω
ω
(2)
Note that Sim
p
(C1,C2) represents similarity of C1’s
parent nodes and C2’s parent nodes and is calculated
via Equation (1).
Sim
so
(C1,C2) represents similarity of C1’s son
nodes and C2’s son nodes. For concept C1 and C2
and their son concepts {C11,C12…,C1i,…C1m}and
{C21,C22…,C2i,…C2n}, we calculated son
similarity with the following steps:
1). For each son concept of C2, calculates the
similarity with each son concept of C1 and records
as Sim(C1i,C2j) ;
2). For 0<j<=n, set Sim(C1i,C2) =
max[Sim(C1i,C2j)];
3). For 0<i<=m, set Sim
so
(C1,C2) =
∑
,
.
Sim
b
(C1,C2) represents similarity of C1’s
brother node and C2’s brother node. The calculation
of Sim
b
(C1,C2) is similar to that of Sim
so
(C1,C2).
ω1, ω2 and ω3 are calculated via the sigmod
function σ(x) =1/(1+e^(-5(x-α)) ) separately (Tang et
al., 2006).
Note that X represents similarity of each measure
and α represents the central point of sigmod
function. In this experiment, we set α=0.5.
3.2.3 Individual Similarity
Each individual in material ontologies represents a
specific material. In general, individual similarity
computing based on the following rules: If the
individuals of two concepts are similar, the concepts
are similar. The individual similarity computing
based on joint distribution is also available for
ontology mapping in material domain (Doan et al.,
2003).
The formula is shown as follows:
Sim
C1, C2
Pc1,c2
P
c1, c2
P
c1
,c2
Pc1,c2
(3)
Figure 3: Ontology mapping process.
Similarity-basedOntologyMappinginMaterialScienceDomain
267
Note that P
c1, c2
represents the probability of
individuals belong to both C1 and C2.
P
c1, c2
represents the probability of individuals
belong to C1 but not to C2. P
c1
,c2
represents the
probability of individuals belong to C2 but not to
C1.
3.2.4 Compositive Similarity
Based on the three calculations of similarity, we
propose the compositive similarity measurement:
SimC1, C2
ω
Si
C1,C2ω
Si
C1,C2 ω
Si
C1,C2
ω
ω
ω
(4)
ω 1, ω2 and ω3 are calculated via the sigmod
function mentioned in section 3.1.2.
3.3 Ontology Mapping
Ontology mapping is the process to create semantic
relations between source ontology and target
ontology. Figure 3 shows the iterative process of
ontology mapping which mainly includes feature
extraction, user interaction, similarity computing and
mapping discovery. Figure 3 shows the mapping
process.
Feature extraction: We utilized Jena API to
handle OWL ontologies and extracted information
such as concepts, individuals and relations from two
heterogeneous material ontologies.
User interaction: Some concept pairs are mapped
manually by users before the mapping process. The
fault mappings will be corrected or deleted after the
mapping process.
Similarity computing: Similarities of concept
pairs are calculated by the compositive measure.
Mapping discovery: The threshold is set to 0.7.
The concept pairs with similarity greater than 0.7
will be mapped.
Iteration: The accuracy of ontology mapping will
be improved by the iteration of the process.
4 EXPERIMENTAL
EVALUATION
According to the mentioned measure, we realized
the transformation from relational databases to
ontologies in material domain in our experiment.
Ontology mapping was done based on the proposed
compositive similarity measure. In this section, the
effect of the compositive measure would be
validated compare to single measure.
Table 1: Ontology information.
Ontology
Name
Concepts Properties Individuals
plastic 26 15 204
rubber 16 15 182
Table 1 shows the information of ontologies
generated from relational databases.
Our development tool is MyEclipse 6.0.1. Jena
API and JWI are used in the program. Part of
compositive similarity is shown in Table 2.
The concept pairs whose compositive similarity
greater than 0.7 would be mapped.
An evaluation criterion is the mapping between
plastic ontology and rubber ontology created
manually. Recall ratio and precision ratio is used to
evaluate experimental results. Recall ratio is defined
as:
Recall =
(5)
Precision ratio is defined as:
Precision =
(6)
Note that both precision ratio and recall ratio of
compositive measure are higher than 90%, far
exceed single measure. The experiment reveals that
compositive measure does improve the uality of
ontology mapping in material domain. Thus it can be
Table 2: Compositive Similarity.
PMMA polybutene rubber polyacrylonitrile
ABS_PMMA 0.9 0.42 0.42 0.22
polybutylene 0.3 0.84 0.56 0.53
polymer 0.33 0.52 0.57 0.42
acrylonitrile 0.27 0.49 0.55 0.72
WEBIST2013-9thInternationalConferenceonWebInformationSystemsandTechnologies
268
Table 3: Experimental Result.
Precision ratio (%) Recall ratio (%)
Name Similarity 44.44 80
Structure Similarity 50 90
Individual Similarity 27.78 100
Compositive Similarity 94.44 94.44
seen that, compositive measure we proposed
performs well in heterogeneous ontology mapping in
material domain.ontology mapping in material
domain.
5 CONCLUSIONS
The experiments show that compositive measure
achieves higher precision ratio and recall ratio
compared to single similarity measure. We can
conclude that compositive similarity measure is
suitable for heterogeneous ontology mapping in
material domain. In addition, our work will be
developed further in the future: In the Ontology
building section ： Rules should be improved in
order to realize automatic mapping between
relational database and OWL ontology in material
domain. In the ontology mapping section：For the
word which is not included in WordNet, word
segmentation should be done before similarity
computing. Firstly, in ontology construction section:
rules should be improved in order to realize
automatic mapping between relational database and
OWL ontology in material domain. Secondly, in the
ontology mapping section: the word which is not
included in WordNet, word segmentation should be
done before similarity computing.
ACKNOWLEDGEMENTS
Supported by the Key Science-Technology Plan of
the National ‘Eleventh Five-Year-Plan’ of China
under Grant No. 2011BAK08B04, the R&D
Infrastructure and Facility Development Program
under Grant No. 2005DKA32800, and the 2012
Ladder Plan Project of Beijing Key Laboratory of
Knowledge Engineering for Materials Science under
Grant No. Z121101002812005.
REFERENCES
Budanitsky A., Hirst G. 2006. Evaluating WordNet-based
Measures of Lexical Semantic Relatedness.
Computational Linguistics, 32(1), 13-47.
Cullot N., Ghawi R., Yétongnon K. 2007. DB2OWL: A
Tool for Automatic Database to Ontology Mapping In:
Ceci et al., eds. Proceedings of the 15th Italian
Symposium on Advanced Database Systems, Torre
Canne, Italy. SEBD, 491-494.
Doan A., Madhavan J., Dhamankar R., Domingos P.,
Halevy A. 2003. Learning to match ontologies on the
Semantic Web. The VLDB Journal, 12(4) pp.303-319.
Ehrig, M., Staab, S. 2004. QOM - Quick Ontology
Mapping In: S. A. Mcllraith, D. Plexousakis, F. V.
Harmelen, eds. 3th International Semantic Web
Conference, Hiroshima, Japan, November 7-11, 2004.
Berlin: Springer Berlin Heidelberg, 683-697.
Euzenat J, Valtchev P．2004. Similarity-based ontology
alignment in OWL-lite In: R. Lopez and L. Saitta, eds.
Proceedings of the European Conference on Artificial
Intelligence, Valencia, Spain, August 22-27, 2004.
Amsterdam: IOS Press, 333-337.
Hu W., Jian N. S., Qu Y. Z., Wang Y. B. 2005. GMO: A
graph matching for ontologies In: Ashpole, Benjamin
and Ehrig et al., eds. Proceedings Of the KCAP 2005
workshop on Integrating Ontologies, Canada, October,
2005. Germany, CEUR-WS.org, 43-50.
Mahapatra L., Mohan M., Khapra M. 2010. OWNS:
Cross-lingual Word Sense Disambiguation Using
Weighted Overlap Counts and Wordnet Based
Similarity Measures In: Proceedings of the 5th
International Workshop on Semantic Evaluation,
Sweden, 2010. PA, USA: Association for
Computational Linguistics Stroudsburg, 138-141.
Tang J., Li J. Z., Liang B. Y. 2006. Using Bayesian
decision for ontology mapping. Journal of Web
Semantics, 4(4), 243-262.
Similarity-basedOntologyMappinginMaterialScienceDomain
269
