=Paper=
{{Paper
|id=None
|storemode=property
|title=Alignment results of SOBOM for OAEI 2010
|pdfUrl=https://ceur-ws.org/Vol-689/oaei10_paper12.pdf
|volume=Vol-689
|dblpUrl=https://dblp.org/rec/conf/semweb/XuWCZ10
}}
==Alignment results of SOBOM for OAEI 2010==
https://ceur-ws.org/Vol-689/oaei10_paper12.pdf
Alignment Results of SOBOM for OAEI 2010
Peigang Xu, Yadong Wang, Liang Cheng, Tianyi Zang
School of Computer Science and Technology
Harbin Institute of Technology, Harbin, China
peigang.xu@gmail.com, ydwang@hit.edu.cn, chl198478@126.com,tianyi.zang@gmail.com
Abstract. In this paper we give a brief explanation of how Sub-Ontology based
Ontology Matching (SOBOM) method gets the alignment results at OAEI2010.
SOBOM deal with an ontology from two different views: an ontology with is-a
hierarchical structure O’ and an ontology with other relationships O’’. Firstly,
from the O’ view, SOBOM starts with a set of anchors provided by a linguistic
matcher. And then it extracts sub-ontologies based on the anchors and ranks
these sub-ontologies according to their depths. Secondly, SOBOM utilizes
Semantic Inductive Similarity Flooding algorithm to compute the similarity of
concepts between different sub-ontologies derived from the two ontologies
according the depth of sub-ontologies to get concept alignments. Finally, from
the O’’ view, SOBOM gets relationship alignments by using the concept
alignment results in O’’. The experiment results show SOBOM can find more
alignment results than other compared relevant methods.
1 System presentation
Currently more and more ontologies are distributedly built and used by different
organizations. And these ontologies are usually light-weighted [1] containing lots of
concepts especially in biomedicine, such as anatomy taxonomy NCI Thesaurus. The
Sub-ontology based Ontology Matching (SOBOM) is designed for matching light-
weight ontologies that has is-a hierarchy as their backbones. It matches an ontology
from two views: O’ and O’’ that are depicted in Fig. 1. The unique feature of our
method is combining sub-ontology extraction with ontology matching.
1.1 State, purpose, general statement
SOBOM is developed to match ontology automatically for general purpose. Based on
two different views, we design three elementary matchers in current version. The first
one is a anchor generator which is used to find anchors; the second one is a structure
matcher SISF (Semantic Inductive Similarity Flooding) which is inspired by Anchor-
Prompt [3] and SF [4] algorithms and is exploited to flood similarity among concepts.
The last one is a relationship matcher which utilizes the results of SISF to get
relationship alignments. In addition, a Sub-ontology Extractor (SoE) is integrated into
SOBOM to extract sub-ontologies according to the anchors got by linguistic matcher
and rank them by their depths descendingly. Overall SOBOM is a sequential method,
so it does not care how to combine the results of different matchers. The overview of
the method is illustrated in Fig. 2.
O O' O ''
Fig. 1. Two views of an ontology in SOBOM
O' O' O' '
Sub_O1
Sub_O2
Sub_O1
Sub_O2
Fig. 2. The processing overview of SOBOM
For simplicity, we define some notations used in the report.
Ontology: An ontology O consists of a set of concepts C, properties (relations) R,
instances I, and axioms A. We use entity e to denote either c ∈ C or r ∈ R . Each
relation r has a domain and range defined as following:
Domain ( r ) = { c i | c i ∈ C and having the relationsh ip r }
Range ( r ) = { c i | c i ∈ C and can be value of r }
Anchor: An anchor is defined as a pair of assumed equivalent non-leaf concepts
across ontologies. Given two ontologies O1 , O2 , c1 ∈ O1 , c2 ∈O2 , if c1 ≡ c2 (means
that c1 is identical with c2),and c1, c2 are both not leaf nodes in the hierarchies of O1
and O2 respectively, then an anchor, X is defined as a pair of concepts < c1 ,c2 > .
Sub-Ontology: Let ontology O = (C,R, HC,HR,). A sub-ontology Ox is a subset of
O whose elements all come from O, called Ox = (C1,H1C), where C1 ⊆ C , H1C ⊆ H C ,
x is the root of HC. Indeed, a sub-ontology in our method is a hierarchical taxonomy,
and its root is an anchor concept.
Sub-ontology Depth. The depth of sub-ontology Ox is the maximal length of path
from the anchor x to the root ri of the taxonomy HiC which contains it in the original
ontology O.
Depth(O x ) = Max( x → ..., → ri ), ri ∈ H iC , H iC ∈ O
1.2 Specific techniques used
SOBOM aims to provide high quality of 1:1 alignments between concept and
property pairs. We have implemented SOBOM algorithm in java and integrated three
distinguishing constitutional matchers. They are independent components in core
matcher library of SOBOM. Due to the space limitation, we only describe the key
features of them. The details can be found in the related paper [8].
z Our anchor generator is based on the local context of an entity in ontology. In
details, the local context of an entity including the following aspects: the
textual information (label, id, comments and so on), the structure information
(the number of super or sub concepts, the number of constraints) and the
individual information (the number of individual if existing). We consider
that the local context of an entity can express the meaning of it. Consequently,
we get three similarity matrixes respectively, and we choose the maximal of
them as the final results.
z SISF uses the RDF statement to represent the ontology and utilizes the
anchors to inducting the construction of similarity propagation graph for the
sub-ontologies. SISF handles the ontology from the view O’ and only
generate concept-concept alignment.
z R-matcher is a relationship matcher base on the definition of the ontology. It
combines the linguistic and semantic information of a relation. From the O’’
view, it utilizes the is-a hierarchy to extend the domain and range of a
relationship and uses the result of SISF to generate the alignment between
relationships.
More importantly, SoE is integrated into SOBOM and extracts sub-ontologies
according to the anchors [5,6]. SoE ranks extracted sub-ontologies according to their
depths. As we extract sub-ontologies for ontology matching, the rules of extracting
sub-ontology in SoE are as following: only sub-concepts of anchor are included in the
sub-ontology. In other words, a sub-ontology is a taxonomy which has anchor as root.
If one of the two concepts in an anchor is a leaf node in the original ontology, we
do not use SISF to deal with it actually. Because this phenomenon just represents a
one-to-many mapping. After extracting sub-ontologies, SOBOM will match these
sub-ontologies according to their depth in original ontology. We first match the sub-
ontologies with larger depth value. By using SoE, SOBOM can reduce the scale of
ontology and make it easy to operate sub-ontologies in SISF.
1.3 Adaptations made for the evaluation
We don’t make any specific adaptation for the tests in the OAEI 2010 campaign. All
the alignments outputted by SOBOM are based on the same set of parameters.
1.4 Link to the system and parameters file
The current version of SOBOM is available at:
http://mlg.hit.edu.cn:8080/Ontology/Download.jsp, and the parameters setting is
illustrated in the reading me file.
1.5 Link to the set of provided alignments (in align format)
We deploy our matcher as a web service, our web service name is:
eu.sealsproject.omt.ws.matcher.AlignmentWSImpl. The endpoint of our web service
can be found at: http://mlg.hit.edu.cn:8080/SOBOMService/SOBOMMatcher?wsdl.
2 Results
In this section, we describe the results of SOBOM algorithm against the benchmark,
directory and anatomy ontologies provided by the OAEI 2010 campaign. We use
Jena-API to parse the RDF and OWL files. The experiments were carried out on a PC
running Windows vista ultimate with Core 2 Duo processors and 4-gigabyte memory.
2.1 Benchmark
On the basis of the nature, we can divide the benchmark dataset into five groups:
#101-104, #201-210, #221-247, #248-266 and #301-304. SOBOM is a sequential
matcher. If the linguistic matcher gets no results, SOBOM will produce no result. We
described the performance of our SOBOM algorithm over each group and overall
performance on the benchmark test set in Table 1.
#101-104 SOBOM plays well for these test cases.
#201-210 In this group, some linguistic features of candidate ontologies are
discarded or modified, their structures are quite similar. SOBOM is a sequential
matcher, our anchor generator matches concepts based on their local context not only
the linguistic information. So, although without linguistic information SOBOM also
gets relatively high precision and recall.
#221-247 The structures of the candidate ontologies are altered in these tests.
However, SOBOM discovers most of the alignments from the linguistic perspective
via our anchor generator, and both the precision and recall are pretty good.
#248-266 Both the linguistic and structural characteristics of the candidate
ontologies are changed heavily, so the tests in this group might be the most difficult
ones in all the benchmark tests. So, SOBOM does not very well.
#301-304 This test group are four real-life ontologies of bibliographic
references. SOBOM can only find equivalence alignment relations.
Table 1. The performance on the benchmark
101-104 201-210 221-247 248-266 301-304
Precision 1.0 0.99 0.99 0.87 0.77
Recall 1.0 0.85 0.99 0.57 0.70
Compared to our previous results (OAEI2009), the recall of every group is highly
improved. This is enhanced by our redesigned anchor generator.
2.2 Anatomy
The anatomy real world test bed covers the domain of body anatomy and consist of
two ontologies, Adult Mouse Anatomy (2247 classes) and NCI Thesaurus (3304
classes). These are relatively large compared to benchmark ontologies. This type
ontologies is what SOBOM suitable for handling, it generated 268 sub-ontologies and
1249 alignments between MA and NCI, consumed 19min3s to complete the matching
task. It is obvious that most of the alignment appears in the leaf nodes in ontologies
(834 leaf node alignments). The experiment result shows in Table 2.
Table 2. The performance of SOBOM on the anatomy test
Leaf node Sub- Total Time
alignments ontologies Alignments consuming
NCI 834 268 1249 19min3s
MA
2.3 Conference
There are 120 pairs of ontologies in this track. Most of them are blind tests (i.e.
there no reference alignment available). The whole results are available at:
http://seals.inrialpes.fr/platform/;jsessionid=FD1E3A5CE8DA43C1D52DB21079EA
ECF3?wicket:bookmarkablePage=:eu.sealsproject.omt.ui.Results&endpoint=http://21
9.217.238.162:8080/SOBOMService/SOBOMMatcher?wsdl&evaluationID=http://21
9.217.238.162:8080/SOBOMService/SOBOMMatcher?wsdl2010/10/03+02:09:03&tr
ack=Conference+Testsuite.
3 General comments
In this section, we want to introduce comments on the results of SOBOM algorithm
and the way to improve it.
3.1 Comments on the results
Strengths SOBOM deals with ontology from two different views and combines
results of every step in a sequential way. If the ontologies have regular literals and
hierarchical structures, SOBOM can achieve satisfactory alignments. And it can avoid
missing alignment in many partitioning matching methods as illustrated in [7].
Weaknesses SOBOM needs anchors to extract sub-ontologies. So it depends on the
precision of anchors. In current version, we use a linguistic matcher to get anchor
concept, if the literals of concept missed, SOBOM will get bad results.
3.2 Discussions on the way to improve the proposed system
SOBOM can be viewed as a frame of ontology matching. So many independent
matchers can be integrated into it. Now, we have enhanced the anchor generator by
not considering the textual information but also the structure information. Our next
plan is to integrate a more powerful matcher to produce anchor concepts or develop a
new method to get anchor concepts. Meanwhile, we plan to develop a mapping
debugging method to refine the results of SOBOM.
4 Conclusion
Ontology matching is very important part of establishing interoperability among
semantic applications. This paper reports our participation in OAEI2010 campaign.
We present the alignment process of SOBOM and describe the specific techniques for
ontology matching. We also show the performance in different alignment tasks. The
strengths and the weaknesses of our proposed approach are summarized and the
possible improvement will be made for the system in the future. We propose a brand
new algorithm to match ontologies. The unique feature of our method is combining
sub-ontology extraction with ontology matching based on two different views of an
ontology.
References
1. Fausto Giunchiglia and Ilya Aihrayeu : Lightweight Ontologies. Technical Report. (2007)
DIT-07-071.
2. G.Stoilos, G. Stamou, S. Kollias: A string metric for ontology alignment, In Proc. Of the 4th
International Semantic Web Conference(ISWC’05). (2005) 623-637.
3. N.F. Noy, M.A. Musen: Anchor-PROMPT: using non-local context for semantic matching,
In Proc. Of IJCAI2001 Workshop on Ontology and Information Sharing, ( 2001) 63-70.
4. S. Melnik, H.G. Molina and E. Rahm: Similarity Flooding: A Versatile Graph Matching
Algorithm, In Proc 18th Int’l Conf. Data Eng. (ECDE’02) (2002) 117-128.
5. Julian Seidenberg and Alan Rector: Web Ontology Segmentation: Analysis, Classification
and Use, WWW2006, (2006).
6. H. Stuckenschmidt and M. Klein: Structure-Based Partitioning of Large Class Hierarchies. In
Proc of the 3rd International Semantic Web Conference (2004).
7. W. Hu, Y. Qu: Block matching for ontologies, In Proc of the 5th International Semantic Web
Conference, LNCS, vol. 4273, Springer (2006) 300-313.
8. P.G. Xu, H.J. Tao: SOBOM: Sub-ontology based Ontology Matching. To be appear.