=Paper=
{{Paper
|id=None
|storemode=property
|title=ServOMap and ServOMap-lt results for OAEI 2012
|pdfUrl=https://ceur-ws.org/Vol-946/oaei12_paper12.pdf
|volume=Vol-946
|dblpUrl=https://dblp.org/rec/conf/semweb/BaD12
}}
==ServOMap and ServOMap-lt results for OAEI 2012
==
https://ceur-ws.org/Vol-946/oaei12_paper12.pdf
ServOMap and ServOMap-lt Results for OAEI 2012
1 1
Mouhamadou Ba , Gayo Diallo
1
LESIM/ISPED, Univ. Bordeaux Segalen, F-33000, France
first.last@isped.u-bordeaux2.fr
Abstract. We present the results obtained by the ontology matching tools
ServOMap and ServOMap-lite within the 8th edition of the Ontology Alignment
Evaluation Initiative (OAEI 2012) campaign. The mappings computation is
based on Information Retrieval techniques thanks to the use of a dynamic
knowledge repository tool, ServO. This is the first participation of the two
systems.
1 Presentation of the systems
We describe in this paper the ServOMap system, a piece of research work related
to the area of ontology matching [1]. The followed matching approach takes its roots
from the Ontology Repository (OR) system ServO [2, 3] and an initial idea
implemented in [4]. The ServO OR provides functionalities for managing multiple
ontologies and providing indexing and searching facilities. Its design is based on the
assumption that there is a real necessity to offer both the possibility of retrieving
online knowledge organization systems (KOS) but also to leverage the many ad hoc
thesauri and other structured vocabularies built and maintained for local purposes.
Indeed, there are many KOS which are not available within the Semantic Web
infrastructure and are not reachable by conventional Semantic Web search engines
and repository (e.g. [5-8]). ServO offers the possibility for an automated and fast OR
building for a particular application purpose. The ServoMap matching system takes
benefit of ServO and is a flexible and efficient large scale ontology matching system.
1.1 Purpose and general statement
ServOMap is designed for facilitating real time interoperability between different
applications which are based on heterogeneous knowledge organization systems. The
heterogeneity comes from the language format, their level of formalism, etc. The
system relies on Information Retrieval (IR) techniques and a dynamic description of
entities of different KOS for computing the similarity between them. It is mainly
designed for meeting the need of matching large scale ontologies such as [9].
From now on, if not necessary, we will mainly continue to refer to ServOMap for
describing our two tools as ServOMap-lt is a version which uses only some of the
settings of the system.
1.2 Techniques used
The overall followed process for matching two inputs ontologies is described in figure 1.
We detail below each step.
Computing Ontology Metrics
The first step after parsing and loading input ontologies is to compute a set of metrics that
are later used as parameters for the systems and for optimization purpose. These metrics
include for any input ontology: the average number of child by concepts, the list of languages
used to denote entities labels or their annotation properties, the most frequent single terms
within the ontology, the longest set of synonyms labels used to describe a concepts.
Fig. 1: ServoMap overall followed process for ontology matching
Lexical and Contextual Indexing
As ServOMap relies on IR techniques for ontology matching, an ontology is seen
as a corpus of document to process where each entity (concepts, relations) is a
semantic document to process.
ServOMap constructs an inverted index thanks to the use of the Ontology
Indexing Module of ServO which relies on the Apache Lucene API1. According to the
parameters computed during the previous step, a dynamic generation of each entity
description is performed. This process is dynamic as each entity is described
according to the features it holds. Therefore, some concepts may have synonyms in
several languages or may have comments, while others may only have English terms.
Moreover, some concepts may have declared properties (either object properties or
data type properties), etc. During this dynamic description process, the retrieved
strings from a concept are passed to a set of filters: stop words removal, normalization
(upper case to lower case), punctuations removal, completion of labels by the
permutations of their terms and so on. A flag is used to indicate whether ServOMap
uses stemming or not and if the words of a term will be concatenated before to add
them to the index. Table 1 gives an extract of available fields and their term counts
within the index for the Foundational Model of Anatomy ontology (FMA). The
version used for this ontology contains 79,042 entities, among them 78,884 are
concepts. As we can see, the value of the dDomain field (the domain of a property) is
spatialassocirelat which is the term “spatial association relation”. And the concept
with id #Accessory_lobar_vein has as directLabelCEn (direct label English label) the
1
http://lucene.apache.org/
set {accessorilobarvein veinaccessorilobar veinlobaraccessori} for “Accessory lobar
vein” and its permutations. All spaces are removed between words.
Term
Field Name Example
Counts
dDomain 15 spatialassocirelat
dRange 5 string
accessorilobarvein
directLabelCEn 152,088 veinaccessorilobar
veinlobaraccessori
directNameC 78,884 accessorilobarvein
directNameP 52 percentag
http://bioontology.org/#Acce
uri 79,042
ssory_lobar_vein
Table 1: An extract of an entry index for the Mouse Anatomy Ontology
Compute lexical based similarity
After the indexing phase, ServOMap proceeds to the computing of lexical based
similarity. This step relies on the Ontology Retrieval Module of the ServO OR.
Depending on the flag indicating the indexed ontologies, the Ontology
Processing Module is called for retrieving the concepts to use for searching over the
built index. Thus, if both input ontologies are indexed, the first one, let’s say O1, is
used as search ontology over the index on the second ontology I2. And, vice versa, the
ontology O2 is used to perform search over the index of the first ontology I1. If the
flag indicates that one ontology is indexed, then ServOMap performs only a one way
search.
As in the lexical and contextual indexing phase, a dynamic generation of entity
description if performed for any entity to use in order to search the index. A Boolean
query is constructed with all the available fields for the entity. Each Boolean query,
represented as a vector of terms, is searched over the index. A ranked list of entities is
retrieved. ServOMap keeps the result constituted by the couple of the entity to search
and the entity having the highest similarity as a possible mapping (vectorial
similarity). It can happen that several entities have the same similarity with the entity
to search. In this case, in order to keep the most relevant one, the names of the entities
are compared using the Levenshtein Distance.
Compute context-based similarity
The idea of context-based similarity is based on the assumption that when two
entities are similar, there is a big chance that the concepts that surround it are also
similar. Here, by surrounding concepts (context) we mean super-concepts, sub-
concepts and siblings concepts. Therefore, in the context based similarity, the
description of a concept is based on its context. This context based similarity is
applied only on concepts and not on the properties of the ontologies to match. In
addition, we restrict the contextual similarity computing to only the concepts that
have not been yet mapped to any other concepts by the lexical-based similarity. This
is based on the assumption that if two concepts are mapped by the previous lexical
strategy, it is likely to be correct.
Refining mappings obtained from context based similarity
The mappings with context similarity are less accurate. The idea is thus to avoid
keeping a couple obtained from the context based similarity where one of the entries
is already mapped during the lexical process by another concept. This strategy takes
into account the worst case and allows removing several incorrect mappings and
increase the recall at the same time. However, it generates false positive
correspondences, and the precision obtained with lexical-based mappings is then
reduced.
Processing disjoints concepts
For ontology matching, some inputs ontologies are described with complex
axioms. In particular, it is possible to have disjointness statements. In such a case, we
use an algorithm for processing these particular issues. Let’s assume that C1 and C2
are two disjoints concepts belonging to an ontology O1 and C3 and C4 two other
disjoints concepts belonging to the ontology O2. During the indexing phase, we
complete the description of C1 by adding a field for its disjoint concepts and the same
for C2, etc. These information is later used to avoid let’s say mapping both C1 – C3 and
C1 – C4.
ServOMAP ServOMap-lt
Terms processing According to the language The same for all languages
of the labels
Entities taken into account All Only Classes
Ontologies indexed Both One
Searching strategy Two ways One way
Stemming No Yes
Arity 1:1 1:n
Table 2: Configurations of ServOMap and ServOMap-lt
1.3 Adaptations made for the evaluation
The ServO OR system uses a threshold as parameter for possibly limiting the
retrieved concepts from the index. For ServOMap we limited the results to the best
similarity.
Our system participated to the campaign with two versions of our approach
corresponding to different parameters settings. The main differences in term of
parameters are presented in table 2.
In addition to these parameters, we used only the first step of similarity computing.
And our system does not use a particular knowledge background.
1.4 Link to the system and parameters file
The Seals wrapped ServoMap and ServOMap tools are available online at
http://code.google.com/p/servo/.
2 Results
In this section, we provide comments on the official results obtained by the two
configurations of the ServOMap matching system.
2.1 benchmark
The Benchmark track 2012 includes 111 tests. Each test concerns a source
ontology called reference and a test ontology which is created by modifying some
information from the reference alignment. For the provided dataset (finance, bench2,
bench3, bench 4 and biblio) ServOMap performed better than ServOMap-lt thanks to
the better recall. Due to the one way searching strategy of ServOMap-lt, it is faster but
its configuration based on stemming and only classes-based strategy reduced its F-
measure.
2.2 anatomy
The precision of our system are very good on the Anatomy track where the
ServOMap configuration provided the best precise mappings (0.996). In term of
computation times, ServoMap-lt completed the task in less than 25 seconds.
2.3 conference
For the conference track, contrary to the results obtained using directly the Seals
Plateform, the official provided results were filtered out by removing all instance-to-
any_entity and owl:Thing-to-any_entity correspondences prior to computing
Precision/Recall/F1-measure. Our system was able completing the 120 alignments in
64 seconds for the ServOMap configuration and in 51 seconds for SevOMap-lt.
2.4 multifarm
Even if our system is able to deal with multilingual ontologies, the cross-lingual
ontology mapping has not yet been implemented, which is the case with the multifarm
task. We were able processing the inputs ontologies but fail computing correct
mappings at this time.
2.5 library
The library track is about matching two thesauri, the STW and the TheSoz
thesaurus. They provide a vocabulary for economic respectively social science
subjects and are used by libraries for indexation and retrieval. As our ontology
processing module relies on the Jena Framework [10], we experienced an issue
processing the input ontologies because of their formatting. However, we were
eventually able completing the task and correctly handled multilingual terminologies
associated with the entities in these KOS. ServOMap-lt and ServOMap were among
the best systems, ranked second and third respectively in term of F-measure (0.670 and
0.665). ServOMap finished the task in 44 seconds (second) and ServOMap-lt in 45
seconds.
2.6 large biomedical ontologies
Our tool in both configurations was able completing the large biomed track
(LargeBio), which was the most challenging one regarding particularly the number of
entities involved in the matching task. We found the NCI thesaurus very time
consuming for context based mapping as its concepts have many siblings. Table 3
summarizes the performances obtained by the ServOMap and ServOMap-lt on the
LargeBio track. ServOMap provided overall the best precision mappings among all
the participating systems (0.903) and completed all the tasks in 2,310 seconds.
ServOMap-lt was ranked second in term of F-measure with 0.780 and completed all
the tasks in 2,405 seconds.
ServOMap ServOMap-lt
P R F T (s) P R F T(s)
0.945 0.747 0.834 327 0.931 0.8 0.86 366
FMA-NCI
FMA- 0.953 0.656 0.777 893 0.956 0.60 0.802 790
SNOMED
SNOMED- 0.901 0.554 0.687 1,089 0.875 0.593 0.706 1,248
NCI
Table 3: Performance obtained on the 2012 LargeBio track
3 General comments
3.1 Comments on the results
Our system performs well for knowledge organization systems having concepts
described by several synonyms terms regardless their languages as it depends heavily
on the lexical description of the resources. However, for the tasks which relies more
on the structural description of ontologies, our system performs less. Overall, the
precision is very good, in particular for the ServOMap configuration as its uses a very
discriminating strategy during the search process (two ways search).
3.2 Discussions on the way to improve the proposed system
So far our system is not using any external resources apart from the usual stops
words list constituted by the common terms discarded during indexing and searching.
It relies only on the intrinsic information encoded into the input ontologies. Our
system could be improved then by the use of external resources for instance for
morphological and lexical variation of terms or by the use of the UMLS and its
semantic network for removing incorrect mappings found during the context-based
similarity. In addition, completing the lexical and contextual description of entities by
true structural information could also improve the results. Also, as ServOMap is not
able to compute oriented mapping, which is quite challenging with an approach
relying on the lexical description of entities, structural description could help. From
computation time point of view, implementing multithreading can be a possible way
to improve the system.
3.3 Comments on the OAEI 2012 procedure
As a first participation, we found the OAEI procedure very convenient and the
organizers very supportive. The use of Seals allows objective assessments.
3.4 Comments on the OAEI 2012 test cases
The OAEI test cases are various and this leads to comparison on different levels of
difficulty, which is very interesting. In addition, real world ontologies are provided.
4 Conclusion
This 2012 edition of OAEI is our first participation in the campaign. The results
obtained both by ServOMap and ServOMap-lt are quite very promising both for F-
measure and computing times. The version of our system which uses the whole
configuration performed less than the lite one on the Large Biomed task in term of F-
measure while it gives the best precision. The lite version is less stable regarding the
others tasks.
Our ontology matching system presents some limitations. And there is a room of
improvements. First, we plan to improve the algorithm used for filtering out the
mappings provided by the context-based matching in order to increase the recall
without reducing the precision. Also, ServOMap does not use any external resource in
the similarity computing process. We intend to use the UMLS resource for better
discarding incorrect mappings for life sciences related ontologies. Moreover, the
current version does not provide oriented mapping nor takes into account matching
two ontologies described in two different languages (e.g. English Vs French). Thus,
an improvement of the system is the implementation of a cross lingual ontology
matching approach and investigating into oriented mappings issue. Finally, we plan
introducing logic assessment of computed mappings [11] and implementing a user
friendly interface.
References
1. Euzenat, J, Meilicke, C, Stuckenschmidt, H, Shvaiko, P, Trojahn, C.: Ontology Alignment
Evaluation Initiative: six years of experience. J Data Semantics (2011)
2. Diallo G. Efficient Building of Local Repository of Distributed Ontologies. In Proceedings
of International Conference on Signal-Image Technology and Internet Based Systems -
SITIS’2012, pp. 159–166. IEEE
3. Diallo G. Towards decentralized and cooperative repositories of distributed ontologies. In
Proceedings of SWAT4LS 2011, pp. 8–9
4. Diallo G, Simonet M, Simonet A. Bringing Together Structured and Unstructured Sources:
The OUMSUIS Approach. OTM Workshops (1) 2006: 699-709
5. Ding, L, Finin, T, Joshi, A, Pan, R, Cost, RS, Peng, Y, Reddivari, P, Doshi, V, Sachs, J
(2004). Swoogle : a search and metadata engine for the semantic web. In Proceedings of the
Thirteenth ACM Conference on Information and Knowledge Management. pages 652-659.
6. Côté RG, Jones P, Apweiler R, Hermjakob H. The Ontology Lookup Service, a lightweight
cross-platform tool for controlled vocabulary queries. BMC Bioinformatics. 2006;7:97
7. d'Aquin, M, Baldassarre, C, Gridinoc, L, Angeletou, S, Sabou, M, Motta, E. Watson: A
Gateway for Next Generation Semantic Web Applications. Poster session of the
International Semantic Web Conference, ISWC 2007.
8. Noy NF, Shah NH, Whetzel PL, Dai B, Dorf M, Griffith N, et al. BioPortal: ontologies and
integrated data resources at the click of a mouse. Nucleic Acids Research. 2009 May 29;37
9. Ruiz EJ, Grau BC, Zhou Y, Horrocks I. Large-scale Interactive Ontology Matching:
Algorithms and Implementation. Proceedings of the 20th European Conference on Artificial
Intelligence (ECAI). IOS Press; 2012. p. 444–9
10.Carroll, JJ, Dickinson, I, Dollin, C, Reynolds, D, Seaborne, A, Wilkinson, K. Jena:
implementing the semantic web recommendations. In Proceedings of the 13th International
World Wide Web Conference, pp. 74-83, New York (2004)
11.Meilicke C, Stuckenschmidt H, Sváb-Zamazal O. A Reasoning-Based Support Tool for
Ontology Mapping Evaluation. ESWC. 2009. p. 878–82