=Paper=
{{Paper
|id=Vol-1545/oaei15_paper16
|storemode=property
|title=STRIM results for OAEI 2015 instance matching evaluation
|pdfUrl=https://ceur-ws.org/Vol-1545/oaei15_paper16.pdf
|volume=Vol-1545
|dblpUrl=https://dblp.org/rec/conf/semweb/KhiatBB15
}}
==STRIM results for OAEI 2015 instance matching evaluation==
https://ceur-ws.org/Vol-1545/oaei15_paper16.pdf
STRIM Results for OAEI 2015 Instance Matching
Evaluation
Abderrahmane Khiat1 , Moussa Benaissa1 and Mohammed Amine Belfedhal2
1
LITIO Laboratory, University of Oran1 Ahmed Ben Bella, Oran, Algeria
abderrahmane khiat@yahoo.com , moussabenaissa@yahoo.fr
2
Evolutionary Engineering and Distributed Information Systems Laboratory (EEDIS), Djillali
LIabes University of Sidi Bel Abbes, Algeria
Mohammed.belfedhal@gmail.com
Abstract. The interest of instance matching grows everyday with the emergence
of linked data. This task is very necessary to interlink semantically data together
in order to be reused and shared. In this paper, we introduce STRIM, an automatic
instance matching tool designed to identify the instances that describe the same
real-world objects. The STRIM system participates for the first time at OAEI
2015 in order to be evaluated and tested. The results of the STRIM system on
instance matching tracks are so far quite promising. In effect, the STRIM system
is the top system on SPIMBENCH tracks.
Keywords: String-Based Similarity, Instance Mapping, Instance Matching, Linked
Data, Web of Data, Semantic Interoperability, Semantic Web.
1 Introduction
The current Web, contains documents in various formats (PDF, Excel, HTML file, etc.)
connected by hypertext links, also known as the Web of Documents. Note that, we mean
by document, if the content is unstructured and not exploitable i.e. the semantic the con-
tent is not presented. Contrary to data, where the content is structured and exploitable
i.e. the semantic of the content is presented using RDF for example.
The inadequacy of the Web of Documents resides in the fact that the content of these
documents is probably unstructured and its semantic is not presented which means that
it is not exploitable and untreatable automatically in different applications, either by
the machine or by expressive queries.
In order to deal with these problems, and especially for the re-use and sharing of
content, the transition from the document to the data is very necessary. This involves
the use of semantic web technologies in order to (a) publish structured data on the Web,
(b) make possible, the links between data from one data source to data within other data
sources. These two points are very important to ensure semantic interoperability.
These data should be expressed using the RDF language (Resource Description
Framework [see section 2.1]) to achieve the two major points that we have mentioned
before in order to enable the semantic interoperability, which led to the emergence of
the Web of Data. The data presented and structure in this form (RDF) can be easily
interpreted by the computer, re-used in applications and easily linked with other data. If
the data are easily linked the computer can work through relationships with other data
and in this case the interoperability will be ensured. Other advantages of Linked Data
among others are: improving the data quality, less human intervention and processing
and short development cycles (quicker and save time).
With the effort of Linked Data Community to publish existing open license datasets
as Linked Data on the Web and interlink things between different data sources, the Web
of Linked Data has seen remarkable increase over the past years. In terms of statistics,
in 2007, over 500 million RDF triples published on the web with around 120,000 RDF
links between data sources. In 2010, over the 28.5 billion triples, in 2011 over 31.6 bil-
lion triples and in 2013 over 50 billion triples. According to these statistics, the Linked
Data seems to be increasing drastically [6].
Linked Data, by definition, links the instances of multiple sources. A common way
to link these instances to others is to use the owl:sameAs property. The enormous vol-
ume of data already available on the web and its continuity to increase, requires tech-
niques and tools capable to identify the instances that describe the same real-world
objects automatically.
With the OAEI evaluation campaign which distinguishes between matching systems
that have participated in the category of ontology matching and those that have partic-
ipated in the category of instance matching, these tools can be tested and evaluated.
However, few systems3 [10] namely InsMT, LogMap and RiMOM-IM have partici-
pated to test their performance at instance matching track of OAEI 2014.
In this paper we deal with two challenges namely:
1. How to link the distributed and heterogeneous data which are described with in-
stances.
2. How to deal with the huge volume of data available on the web and its continuity
to increase [14].
Indeed, the Solution to this problem consists to provide techniques and tools capable
to identify the instances that describe the same real-world objects automatically.
In this paper, we describe the STRIM system in order to resolve automatically the
instance matching problem. The STRIM system, extracts first all information about the
two instances to be matched and normalizes them using NLP techniques. Then, it ap-
plies edit distance as a matcher to calculate the similarities between the normalized
information. Finally, the approach selects the equivalent instances based on the maxi-
mum of shared information between the two instances.
The STRIM system has participated for the first time at OAEI evaluation campaign
and it provides very good results in terms of precision, recall and f-measure.
The rest of the paper is organized as follows. First, the preliminaries on instance
matching are presented in section 2. In the Section 3, we presented the related work
on instance matching systems that participated in Instance Matching Track of OAEI
2014. In the Section 4, we describe our system by giving a detailed account of our
approach. The experimentation results is presented in Section 5. The Section 6 contains
concluding remarks and sets directions for future work.
3
The declaration of OAEI 2014 evaluation campaign about instance matching systems Again,
given the high number of publications on data interlinking, it is surprising to have so few
participants to the instance matching track, although this number has increased.
2 Preliminaries
In this section, we present the basic notions related to Instance Matching by explaining
the linked data and instance matching definition.
2.1 Linked Data Principle
The Linked Data consist to relate data with typed links across the Web using URIs,
HTTP and RDF. The linked Data principles are defined by Tim Berners-Lee in 2007
[11]. These principles are as follow:
– Use URIs as names for things.
– Use HTTP URIs so that people can look up those names.
– When someone looks up a URI, provide useful RDF information.
– Include RDF statements that link to other URIs so that they can discover related
things.
Fig. 1: Linked Data
The Linked Data (Fig. 1), by definition [12], links the instances of multiple sources.
A common way to link the instances in these sources to others, is the use of the owl:sameAs
property. Instance matching is required to interlink these data.
2.2 RDF Language
These data should be expressed using the RDF language (Resource Description Frame-
work) to achieve the two major points that we have mentioned before in order to enable
the semantic interoperability. The RDF language is a graph model to formally describe
Web resources and metadata, in order to allow automatic processing of such descrip-
tions [13][1][2]. An RDF file thus formed is a labeled directed multi-graph. Each triplet
corresponds to a directed arc whose label is the predicate, the source node is the subject
and the target node is the object.
2.3 Instance Matching Definition
The Instance Matching (Fig.2) is a process that starts from collections of data as input
and produces a set of mappings (simple or complex) between entities of the collections
as output [5].
Fig. 2: Instance Matching Process
2.4 Entity Resolution Notion
Definition [5]: Let D1 and D2 be represent two datasets, each one contains a set of data
individuals Ti which are structured according to a schema Oi . Each individual Ii j Ti
describes some entity wj .
Two individuals are said to be equivalent Ij Ik if they describe the same entity wj
= wk according to a chosen identity criterion. The goal of the entity resolution task is
to discover all pairs of individuals (I1 i, I2 j) — I1 i T1 , I2 j T2 such that w1 i = w2 j.
In the context of linked data, datasets Di are represented by RDF graphs. Individuals
Ii Ti are identified by URIs and described using the classification schema and properties
defined in the corresponding ontology Oi .
Example of Instance Matching We give below an example that shows how to link
data from DBpedia with other data sources using the owl:sameAs property.
owl:sameAs
3 Related Work
We present and discuss in this section the major works relevant to instance matching
that participated at OAEI 2014 evaluation campaign. Only two systems succeed to finish
all sub-tracks of instance matching track of OAEI 2014, namely RiMOM-IM and our
previous InsMT system. We cite in exhaustive way only the instance matching systems
that have participated in OAEI 2014 evaluation campaign and which are the object of
comparison with our system STRIM.
1) LogMap [7]: The LogMap family participated with four different versions namely
LogMap, LogMap-Bio, LogMap-C and LogMapLite in OAEI 2014. Only two versions
(LogMap and LogMap-C) of them have participated at instance matching track. The
LogMap-family system is a highly scalable ontology matching system with built-in rea-
soning and inconsistency repair capabilities. The two versions of LogMap systems iden-
tifies mappings between instances. The LogMap and LogMap-C systems finish only the
first sub-track of instance matching of OAEI 2014 which is Identity Recognition.
2) RiMOM-IM [9] [3] [4]: is an acronym of Risk Minimization based Ontology
Mapping Instance Matching. The principle of RiMOM-IM is to construct a document
from the dataset by extracting the instances information. Then, it uses cosine-similarity
to compare documents. The version of RiMOM-IM system that participated in OAEI
2014 for instance matching is developed based on ontology matching system RiMOM
with some changes in objective. The objective of RiMoM-IM is to solve the challenges
in large-scale instance matching by proposing a novel blocking method.
3) InsMT(L) [8]: is an acronym of Instance Matching at Terminological (Linguis-
tic) level. InsMT(L) has participated for the first time in OAEI 2014. The principle
of InsMT(L) is to use String-based algorithms (and WordNet as matcher at linguistic
level) in order to calculate similarities between instances after the annotation step. The
similarities calculated by each matcher are aggregated using the average aggregation
strategy after a local filtering. Finally InsMT(L) system operates a global filtering in
order to identify the alignment. The InsMT(L) system shows good results in terms of
recall on different sub-tracks of instance matching of OAEI 2014. The InsMT(L) system
finishes all sub-tracks of instance matching of OAEI 2014 which is Identity Recognition
and Similarity Recognition.
4) Other Approaches:
There are several other instance matching approaches like HMatch [18], FBEM
[17], SILK [16] and the works proposed in [15] which are not covered by this paper due
to minor importance for our approach. These instance matching approaches have not
participated in instance matching track of OAEI 2014. With respect to these approaches,
we did not take them in consideration because we do not have their official results for
the experimental protocol of OAEI in 2014.
As we have mentioned before, with the high number of publications about interlink-
ing approaches only a few systems have participated at OAEI 2014. These systems are
LogMap, RiMoM-IM and our previous InsMT(L) system.
4 STRIM: STRing based algorithm for Instance Matching
We summarize the process of our approach to provide a general idea of the proposed
solution. It consists in the following successive phases:
4.1 Extraction and Normalization
The system extracts from each individual Ii P1 m1 ; P2 m2 ,... a set of information m1 ,
m2 , ... using different properties P1 , P2 , .... Then, NLP techniques are applied to nor-
malize these infrmation. In particular, three pre-processing steps are performed: (1)
case conversion (conversion of all words in same upper or lower case) (2) lemmatiza-
tion stemming and (3) stop word elimination. Since String based algorithm is used to
calculate the similarities between information, these steps are necessary.
4.2 Similarity Calculation
In this step, the system calculates the similarities between the normalized informations
using edit distance as string matcher. Our system selects the maximum similarity values
calculated between different informations by edit distance. If two informations are the
same (based on maximim similarity values) the counter is incremented to 1, etc.
4.3 Identification
Finally, we apply a filter on maximum counter values in order to select the correspon-
dences which mean that the selected correspondences (equivalent individuals) are those
who share maximum informations.
5 Experimentation
In this section, we present the results (Tab. 1) obtained by running our STRIM system
on instance matching tracks of OAEI 2015 evaluation campaign.
Table 1: The Results of STRIM System
System Track Precision F-measure Recall
STRIM sandbox val-sem task 0.90 0.95 0.99
LogMap sandbox val-sem task 0.99 0.92 0.86
STRIM mainbox val-sem task 0.91 0.95 0.99
LogMap mainbox val-sem task 0.99 0.92 0.85
STRIM sandbox val-struct task 0.99 0.99 0.99
LogMap sandbox val-struct task 0.99 0.90 0.82
STRIM mainbox val-struct task 0.99 0.99 0.99
LogMap mainbox val-struct task 0.99 0.90 0.82
STRIM sandbox val-struct-sem task 0.91 0.95 0.99
LogMap sandbox val-struct-sem task 0.99 0.88 0.79
STRIM mainbox val-struct-sem task 0.91 0.95 0.99
LogMap mainbox val-struct-sem task 0.99 0.88 0.79
Only two systems have participated at SPIMBNNCH tracks namely the LogMap
and STRIM systems. The SPIMBENCH consists of the following three different tasks:
val-sem, val-struct and val-sem-struct. Each task has two tests (1) the Sandbox which
contains two datasets in small scale and (2) the Mainbox which contains two datasets in
large scale. The goal of three tasks consists to determine when two OWL instances de-
scribe the same Creative Work. However, the three tasks have been produced by altering
a set of original data. In other words, the datasets of the val-sem task have been pro-
duced by using value-based and semantics-aware transformations. For the datasets of
the val-struct task have been produced by using value-based and structure-based trans-
formations. Finally the datasets of the val-sem-struct task have been produced by using
value-based, structure-based and semantics-aware transformations.
We have evaluated the results of STRIM system based on the results obtained on
Mainbox tests. The reason is that these tests were blind (i.e. the reference alignment
is not given to the participants) during the evaluation of Instance matching systems by
the OAEI evaluation campaign. On the other side, in the Sandbox tests, the reference
alignment were available to help the instance matching systems to configure theirs pa-
rameters.
Regarding F-measure results, the STRIM system seems to achieve the best results
before the LogMap system. The F-measure is always more than 95%. we can remark
that STRIM system achieve high recall for the three tasks. It always equal to 99%.
* As conclusion, the result proves that our STRIM system is effective and efficient
for the three tasks of SPIMBENCH track of OAEI 2015.
6 Conclusion
In this article, we have introduced STRIM, an instance matching system which consists
in identifying the instances that describe the same real-world objects automatically. Our
approach is useful, especially when the instances contain terminological information.
The STRIM system is composed of three steps: the first step consists in extracting
and normalizing all information about the two instances to be matched. The second step
consists in applying an edit distance as a matcher to calculate the similarities between
the normalized information. The final step, consists in selecting the equivalent instances
based on the maximum of shared information between the two instances.
The STRIM system has participated for the first time at OAEI evaluation campaign
and it provides very good results in terms of precision, f-measure and recall at Instance
Matching of OAEI 2015.
As future perspective, we attempt to apply STRIM to link data on could computing
environment and develop other approaches.
References
1. J. Euzenat and P. Shvaiko Ontology Matching, Second Edition, Springer-Verlag, Heidelberg,
pp. 1-511, 2013.
2. M. Ehrig Ontology Alignment: Bridging the Semantic Gap, Semantic Web And Beyond
Computing for Human Experience 4, Springer, pp. 1-250, 2007.
3. Z. Wang, X. Zhang, L. Hou,Y. Zhao, J. Li, Y. Qi and J. Tang RiMOM results for OAEI 2010,
In The Proceedings of the 4th International Workshop on Ontology Matching co-located with
the 9th International Semantic Web Conference (ISWC 2010), pp. 195-202. CEUR-WS.org,
Vol. 689, Shanghai, China, 2010.
4. J. Li, J. Tang, Y. Li and Q. Luo RiMoM: a Dynamic Multistrategy Ontology Alignment
Framework, Journal IEEE Transactions on Knowledge and Data Engineering, vol. 21, No. 8,
pp. 1218-1232, 2009.
5. A. Ferrara, A. Nikolov, J. Noessner and F. Scharffe Evaluation of instance matching tools:
The experience of OAEI, Journal of Web Semantics: Science, Services and Agents on the
World Wide Web, Vol. 21 pp. 49-60, 2013.
6. C. Bizer, T. Heath, and T. Berners-Lee. Linked Data - The Story So Far. International Journal
on Semantic Web and Information Systems (IJSWIS), 2009.
7. E. Jimnez-Ruiz, B. C. Grau, W Xia, A. Solimando, X. Chen, V. Cross, Y. Gong, S. Zhang
and A. Chennai-Thiagarajan LogMap family results for OAEI 2014. In Proceedings of the
9th International Workshop on Ontology Matching co-located with the 13th International
Semantic Web Conference (ISWC 2014), October 20, pp. 126-134. CEUR-WS.org, Trentino,
Italy, 2014.
8. A. Khiat, M. Benaissa, InsMT / InsMTL results for OAEI 2014 instance matching. In Pro-
ceedings of the 9th International Workshop on Ontology Matching co-located with the 13th
International Semantic Web Conference (ISWC 2014), October 20, pp. 120-125. CEUR-
WS.org, Trentino, Italy, 2014.
9. C. Shao, L. Hu and J. Li, RiMOM-IM results for OAEI 2014. In Proceedings of the 9th Inter-
national Workshop on Ontology Matching co-located with the 13th International Semantic
Web Conference (ISWC 2014), October 20, pp. 149-154. CEUR-WS.org, Trentino, Italy,
2014.
10. Z. Dragisic, K. Eckert, J. Euzenat, D. Faria, A. Ferrara, R. Granada, V. Ivanova, E. Jimnez-
Ruiz, A. O. Kempf, P. Lambrix, S. Montanelli, H. Paulheim, D. Ritze, P. Shvaiko, A. Soli-
mando, C. Trojahn, O. Zamazal, B. C. Grau, Results of the Ontology Alignment Evaluation
Initiative 2014. In Proceedings of the 9th International Workshop on Ontology Matching co-
located with the 13th International Semantic Web Conference, pp. 61-104. CEUR-WS.org,
Trentino, Italy, 2014.
11. Tim Berners-Lee. Linked Data - Design Issues, 2006.
http://www.w3.org/DesignIssues/LinkedData.html. 7, 26, 82.
12. R. Parundekar, C.A. Knoblock, J.L. Ambite, Linking and building ontologies of linked data.
In: Proceedings of the 9th International Semantic Web Conference (ISWC 2010). Shanghai,
China, 2010.
13. G. Klyne and J. J. Carroll. Resource Description Framework (RDF): Concepts and Abstract
Syntax - W3C Recommendation, http://www.w3.org/TR/rdf-concepts/, 2004.
14. P. Shvaiko and J. Euzenat. Ten challenges for ontology matching. In R. Meersman and Z.
Tari, editors, On the Move to Meaningful Internet Systems: OTM 2008, volume 5332 of
Lecture Notes in Computer Science, pp. 11641182. 2008.
15. D. Engmann and S. Mamann. Instance matching with COMA++. In Proceedings of Daten-
banksysteme in Business, Technologie and Web(BTW 07), pages 2837, 2007.
16. J. Volz, C. Bizer, M. Gaedke, and G. Kobilarov. Discovering and maintaining links on the
web of data, In The Proceedings of 8th International Semantic Web Conference (ISWC
2009), A. Bernstein, D. Karger, T. Heath, L. Feigenbaum, D. Maynard, E. Motta, and K.
Thirunarayan, editors, The Semantic Web - ISWC 2009, volume 5823 of Lecture Notes in
Computer Science, pp. 650665. Springer Berlin / Heidelberg, 2009.
17. H. Stoermer and N. Rassadko. Results of okkam feature based entity matching algorithm for
instance matching contest of oaei 2009, 2009.
18. S. Castano, A. Ferrara, S. Montanelli, and D. Lorusso. Instance matching for ontology popu-
lation. In Proceedings of the 16th Italian Symposium on Advanced Database Systems, pages
121132, 2008.