=Paper=
{{Paper
|id=Vol-1545/oaei15_paper7
|storemode=property
|title=InsMT+ results for OAEI 2015 instance matching
|pdfUrl=https://ceur-ws.org/Vol-1545/oaei15_paper7.pdf
|volume=Vol-1545
|dblpUrl=https://dblp.org/rec/conf/semweb/KhiatB15
}}
==InsMT+ results for OAEI 2015 instance matching==
https://ceur-ws.org/Vol-1545/oaei15_paper7.pdf
InsMT+ Results for OAEI 2015 Instance Matching
Abderrahmane Khiat1 , Moussa Benaissa1
LITIO Laboratory, University of Oran1 Ahmed Ben Bella, Oran, Algeria
abderrahmane khiat@yahoo.com , moussabenaissa@yahoo.fr
Abstract. The InsMT+ is an improved version of InsMT system participated
at OAEI 2014. The InsMT+ an automatic instance matching system which con-
sists in identifying the instances that describe the same real-world objects. The
InsMT+ applies different string-based matchers with a local filter. This is the
second participation of our system and we have improved somehow the results
obtained by the previous version.
Keywords: Terminological Techniques, String Based Similarity, Instance Map-
ping, Instance Matching, Linked Data, Web of Data, Semantic Interoperability,
Semantic Web.
1 Presentation of the System
1.1 State, Purpose, General Statement
The objective of Linked Data with the emergence of the Web of Data is to interlink
semantically data together in order to be reused and processed automatically by the
software agents. These data described by instances are heterogeneous and distributed.
The Instance matching is a very necessary task in Linked Data; it aims to identify the
instances that describe the same real-world objects.
The enormous volume of data already available on the web and its continuity to
increase, requires techniques and tools capable to identify the instances that describe
the same real-world objects automatically.
In this paper, we describe InsMT+ an improved version of our InsMT system which
participated in OAEI 2014. This second version consists to apply different string-based
matchers with a local filter. The second version shows good results better than the
previous one but still not very satisfiable. The details of each step of our system are
described in the following section.
1.2 Specific Techniques Used
The process of our system consists in the following successive steps.
Step 1: Extraction and Normalization of Instances In this step, our system extracts
the instances. Then, we have applied (1) case conversion (conversion of all words in
same upper or lower case) and (2) stop word elimination to normalize the instance
informations.
Step 2: Terminological Matchers In this step, our system calculates the similarities
between instances, normalized in previous phase, using various string-based match-
ing algorithms. More precisely the different string-based matching algorithms used
are: levenshtein-distance, Jaro, SLIM-Winkler. The calculations of similarities by each
string matching algorithm are represented in matrix.
Step 3: Local Filter In this step, our system applies a local filter on each matrix i.e.
we choose for each string-based matching algorithm a threshold to realize a filter. We
consider that: the similarities which are less than the threshold are set to 0. Our intu-
ition behind this local filter is that the similarities which are less than the threshold can
influence the strategy of the average aggregation.
Step 4: Aggregation of Similarities In this step, our system combines the similari-
ties of each matrix (after we have applied a local filter) using the average aggregation
method and the result of the aggregation is represented in a matrix.
Step 5: Global Filter and Identification of Alignment In this step, our system applies
a second filter on the combined matrix (result of the previous step) in order to select the
correspondences found using the maximum strategy with a threshold.
1.3 Adaptations Made for the Evaluation
We do not have made any specific adaptation for this first version of InsMT+, for OAEI
2015 evaluation campaign. All parameters are the same for instance matching track of
OAEI 2015.
1.4 Link to the set of provided alignments (in align format)
The result of InsMT+ system can be downloaded from OAEI 2015 website http:
//islab.di.unimi.it/im_oaei_2015/index.html
2 Results
In this section, we present the results obtained by running InsMT+ on instance matching
track of OAEI 2015 evaluation campaign.
2.1 Author Disambiguation Task
The goal of the author-dis task is to link OWL instances referring to the same person
(i.e., author) based on their publications.
We present below the results obtained by running InsMT+ system on author disam-
biguation task (see Tab. 1).
Table 1: The results of InsMT+ on the Author Disambiguation Task of OAEI 2015.
Track System Expected mappings Retrieved mappings Precision Recall F-measure
Sandbox task EXONA 854 854 0.941 0.941 0.941
Mainbox task EXONA 8428 144827 0.0 0.0 NaN
Sandbox task InsMT+ 854 722 0.834 0.705 0.764
Mainbox task InsMT+ 8428 7372 0.76 0.665 0.709
Sandbox task Lily 854 854 0.981 0.981 0.981
Mainbox task Lily 8428 8428 0.964 0.964 0.964
Sandbox task LogMap 854 779 0.994 0.906 0.948
Mainbox task LogMap 8428 7030 0.996 0.831 0.906
Sandbox task RiMOM 854 854 0.929 0.929 0.929
Mainbox task RiMOM 8428 8428 0.911 0.911 0.911
* The results of InsMT+ are better compared to the first version participated in
OAEI 2014, we can say that we have improved the results in terms of precision. How-
ever, the results are less better than other systems due to the simple techniques used in
InsMT+. Since, InsMT+ is based only on String-based similarity.
2.2 Author Recognition Task
The goal of the author-rec task is to associate a person (i.e., author) with the correspond-
ing publication report containing aggregated information about the publication activity
of the person, such as number of publications, h-index, years of activity, number of
citations.
We present below the results obtained by running InsMT+ system on author recog-
nition task (see Tab. 2).
Table 2: The results of InsMT+ on the Author Recognition Task of OAEI 2015.
Track System Expected mappings Retrieved mappings Precision Recall F-measure
Sandbox task EXONA 854 854 0.518 0.518 0.518
Mainbox task EXONA 8428 8428 0.409 0.409 0.409
Sandbox task InsMT+ 854 90 0.556 0.059 0.106
Mainbox task InsMT+ 8428 961 0.246 0.028 0.05
Sandbox task Lily 854 854 1.0 1.0 1.0
Mainbox task Lily 8428 8424 0.999 0.998 0.999
Sandbox task LogMap 854 854 1.0 1.0 1.0
Mainbox task LogMap 8428 8436 0.999 1.0 0.999
Sandbox task RiMOM 854 854 1.0 1.0 1.0
Mainbox task RiMOM 8428 8428 0.999 0.999 0.999
* The results of InsMT+ on this track are not at all very satisfiable. However, we
can remark that the number of retrieved mappings by our system is less 10 time than
the mappings discovered by other systems, which explained the results obtained. We
are trying to analyses the reason of these results in order to improve our system.
3 Conclusion
This is the second time that InsMT+ system has participated in SEAL platform and
OAEI campaign. In this year, our system has participated only in two instance matching
tracks of OAEI 2015 evaluation campaign. The InsMT+ system gives good results better
than the InsMT system but these results still not statifaisable. As future Perspective, we
attempt to improve more our system in order to get better results.
References
1. A. Doan, J. Madhavan, P. Domingos, and A. Halevy, Learning to map ontologies on the
semantic web, in Proceedings of the International World Wide Web Conference (2003).
2. A. Maedche and V. Zacharias, Clustering ontologybased metadata in the semantic web, in
Proceedings of the 13th ECML and 6th PKDD, (2002).
3. A. Khiat, M. Benaissa, InsMT / InsMTL results for OAEI 2014 instance matching. In
Proceedings of the 9th International Workshop on Ontology Matching co-located with
the 13th International Semantic Web Conference (ISWC 2014), October 20, pp. 120-125.
CEURWS.org, Trentino, Italy, 2014.
4. A. Maedche, B. Motik, N. Silva and R. Volz ”Mafraa mappingframework for distributed
ontologies”, Springer, Benjamins VR (eds) EKAW, Berlin, vol 2473, pp 235250, (2002).
5. K. Todorov, P. Geibel, KU. Kuhnberger ”Mining concept similarities for heterogeneous on-
tologies”, Springer, Berlin, ICDM, vol 6171. , pp 86100, (2010).
6. J. Euzenat and P. Valtchev, Similarity-based ontology alignment in owllite, in Proceedings
of ECAI, (2004).
7. J. Euzenat and P. Shvaiko. OntologyMatching. Springer (2007).
8. M. Ehrig. Ontology Alignment Bridging the Semantic Gap. Springer (2007).
9. M. Jaro. Advances in record-linkage methodology as applied to matching the 1985 census
of tampa, florida. Journal of America Statistical Association, 84(406):414-420, (1989).
10. A. Khiat et M. Benaissa: ”Nouvelle Approche dAlignement dOntologies base dInstances
: trasferet des instances par linfrence”, In The Proceeding of International Conference On
Artificial Intelligence and Information Technology, ICA2IT 2014, Ouargla, Algeria, (2014).
11. V. Levenshtein. Binary codes capable of correcting deletions, insertions, and reversals. Soviet
Physics Doklady, 10:707-710, (1966).
12. W. Winkler. The state of record linkage and current research problems. Statistics of Income
Division, Internal Revenue Service. Publication R99/04 (1999).
13. M. Ehrig and Y. Sure, Ontology mapping - an integrated approach, in Proceedings of the
European Semantic Web Symposium ESWS, (2004).
14. B. Schopman, S. Wang, A. Isaac and S. Schlobach, Instance-Based Ontology Alignment by
Instance Enrichment, Journal on Data Semantics, vol. 1, N 4, (2012).
15. E. Rahm Towards large-scale schema and ontology Alignment, ReCALL, (2011).
16. J. Li, J. Tang, Y. Li and Q. Luo, Rimom: a dynamic multistrategy ontology alignment frame-
work, IEEE Trans Knowl, (2009).