=Paper=
{{Paper
|id=Vol-1230/paper-05
|storemode=property
|title=Enhancing Alignment Results in Ontology Matching for Smart Cities
|pdfUrl=https://ceur-ws.org/Vol-1230/paper-05.pdf
|volume=Vol-1230
|dblpUrl=https://dblp.org/rec/conf/bir/Otero-CerdeiraRG14
}}
==Enhancing Alignment Results in Ontology Matching for Smart Cities==
https://ceur-ws.org/Vol-1230/paper-05.pdf
2nd International Workshop on Ontologies and Information Systems
Enhancing Alignment Results in Ontology Matching for
Smart Cities
Lorena Otero-Cerdeira, Francisco J. Rodríguez-Martínez and Alma Gómez-
Rodríguez
LIA2 Research Group. Computer Science Department. University of Vigo. Spain
{locerdeira, franjrm, alma}@uvigo.es
Abstract. In this paper we propose the use of an ontology matching algorithm to
guarantee the interoperability of the different agents that integrate an smart city.
In this sort of environment the different parties need to cooperate and to integrate
their information in order to provide enhanced services to the users of the smart
city. As the information of these parties may be described by means of different
and heterogeneous ontologies, we find the solution in the use of ontology
matching techniques. The algorithm presented was designed to be able to exploit
the knowledge of previous matched agents to enhance the results and provide the
most accurate results possible.
Keywords: internet of things, smart cities, ontology, ontology matching,
alignment reuse
1 Introduction
In the last years there has been a remarkable increase in the amount of projects and
initiatives related to Internet of Things [1] and Smart Cities [2]. The Internet of Things
is the evolution of the information and communication technologies (ICT), that is
taking us from having connectivity at anytime and anyplace to also having it with
anything. This situation is reflected by the growing amount of different devices with
connecting capabilities, such as RFID tags, NFC devices, sensors, actuators, etc. Such
devices are the building blocks of the smart cities.
The idea behind integrating these devices in a city is to turn it into a smart one, so
citizen’s lives can be improved with new types of services and comfort. These services
can be related to almost every aspect of city life and infrastructure, water and energy
supply, transportation, healthcare, education, etc. [23], and precisely the cities are
looking at ICT to offer services to citizens while reducing costs and improving
efficiencies.
To turn a city into a smart one, the first task to address is to develop a rich
environment of networks that support digital applications [2]. This task involves,
firstly deploying the proper infrastructure which includes different types of sensors,
smart devices and actuators, together with the actual networks that allow the
communication of these. However, the devices by themselves are not enough and it is
necessary to develop applications that exploit these networks of devices.
55
2nd International Workshop on Ontologies and Information Systems
Hence, in a smart city, smart devices, Sensor Networks (SNs) [23] and applications
to exploit them, assume a crucial role. These urban sensors and sensor networks are
generally spread over a wide area and continuously measuring different variables. The
data collected is processed by the different applications which may trigger an action in
some actuator or the response to a user’s request.
It is highly likely that the deployment of a smart city is not done all at once but in a
series of steps, so it is equally likely that different parts of the smart city are developed
by different parties, resulting in the coexistence of different public and private
deployments, each one of which possibly using different smart devices and also their
own hardware and software architectures. It is necessary to guarantee the success of a
smart city to put a special interest in allowing that these different deployments will be
able to interact and seamlessly communicate with each other and, that the information
gathered by the different devices will be properly integrated and shared among the
different systems [26].
This problem is not new to the research community and several alternatives have
been already proposed [26][25][11]. These approaches propose using a wrap for the
different sensors, or compel the use of some standard or protocol to allow the
communication between parties with different knowledge representations. Other
efforts include the use of ontologies to semantically describe services and devices
available [3]. The work that we have developed is in line with the latest but what we
propose exploits ontologies differently.
Our proposal includes the use of ontology matching techniques [4] to guarantee the
connectivity among the different parties in a smart city.
The remaining of the paper is organized as follows. In section 2, we delve into the
use of ontology matching in smart cities and provide the foundations that supported the
development of our system. In section 3, a description of our solution is provided and
discussed. Finally in section 4, the main conclusions and future lines of work are
summarized.
2 Ontology Matching in Smart Cities
A smart city may be seen as a distributed system where several agents on behalf of
their users collect data from the environment by using different sensors. The concept of
user here should be globally understood, as the user of an agent may be a citizen, a
smart device, an application, another agent, etc. The use of ontologies in smart cities is
not new as there is for instance the SCRIBE ontology [24] [5] designed out of the
information gathered from different cities or the ܱܵ ܣܫܨplatform [6]. Ontologies help
in providing a vocabulary to describe a certain domain and the specification of the
meaning of the terms in that vocabulary [4], in our concrete case, ontologies help in
defining the different events, entities and services in a smart city. Besides, they are
particularly suitable for describing the meaning of the concepts in a communication
process between the agents in a Multi-agent System (MAS) [7] and hence they are
used as a way of reducing the semantic gap among the different interacting parties.
56
2nd International Workshop on Ontologies and Information Systems
Fig. 1. Fig. 2. Classification of matching techniques
However, there are several reasons why ontologies by themselves are not enough to
guarantee the interoperability of the different agents. For instance, the agents may use
different ontologies to represent the information gathered from the sensors, the
software applications in the smart city may be developed by different providers that
57
2nd International Workshop on Ontologies and Information Systems
represent their internal knowledge using different ontologies, there may be agents or
applications included in the smart city in a later stage or even itinerant agents that only
need a concrete service at a certain time. In order to actually reduce the heterogeneity
in the definitions and allow a seamless communication of the parties, we relied on
ontology matching techniques [8].
These techniques allow the identification of alignments for pairs of ontologies
where an alignment identifies the set of correspondences holding between the entities
belonging to the ontologies [9]. Apart from the manual identification of
correspondences fulfilled by human experts which has been practically dismissed due
to its cost, there are automatic and semi-automatic methods to compute the alignment
between the ontologies which exploit different features of the ontologies or use
external resources to identify the possible correspondences between the concepts.
Different classifications have been made for the matching techniques although for
the scope of this paper, we followed the one that Euzenat and Shvaiko propose in [4].
This classification, as shown in figure 119, can be read both top-down, then stressing
the interpretation that the different techniques provide for the input information, and
also bottom-up, focussing on the type of input that the matching techniques use.
Regardless of the direction of the reading, they both meet at the concrete techniques
layer.
In the following section, while describing our solution to the ontology matching
problem in smart cities, we briefly describe the different techniques that we have used
linking them to this classification.
3 Solution Description
In this section we briefly describe our algorithm for ontology matching in smart
cities and how we have enhanced its results by following an alignment reuse [12]
approach.
It takes as input two OWL [18] ontologies and relies on the exploitation of some
initial correspondences which we named binding points and which are similar to the
anchors initially used by systems such as LogMap [13], Anchor-Flood [14], Anchor-
Prompt [15] or ASCO [16], although the procedure followed to compute the binding
points is remarkably different to the one used to obtain the anchors in each one of these
systems.
These initial correspondences are obtained by using some language-based and
terminological techniques. The language-based techniques consider names as words in
a natural language and exploit their morphological features. Some of the methods used,
as part of the pre-processing of the strings, are tokenisation, that consists of splitting
words into shorter sequences by means of a separator (blanks, punctuation marks,
camel-case changes, etc) and stopword elimination, that consists of removing words
such as articles, prepositions, etc.
On the other hand, the terminological techniques consider their inputs just as strings
and apply string-distance measures to asses the similarity between two entities. In our
case we have used Jaro-Winkler distance [10] and Levenshtein distance [10] on the
19 Extracted from the book Ontology Matching [4]
58
2nd International Workshop on Ontologies and Information Systems
pre-processed strings. The results of these distances are weighted in order to obtain an
only lexical value for each pair of entities in the ontologies to match. To weight the
results of these measures, another similarity distance is used, in this case, it is based on
the exploitation of WordNet [17] as a external resource. This is also a language-based
technique that takes advantage of the definitions provided by this lexical database to
evaluate the distance between two terms.
Once the similarity between the terms in the ontologies has been determined, only
those pairs with the highest value are selected to become the initial binding points.
These initial correspondences sequentially undergo several procedures that take
advantage of some structural features of the ontologies and that allow the discovery of
new binding points. These binding points can identify both pairs of classes or
properties. Each one of the newly discovered binding points is assigned a tag that
identifies the procedure and branch within it that led to its discovery. If a binding point
is reached by several procedures, all the tags are recorded.
1. Properties Inverse Procedure: this procedure retrieves new correspondences
between properties by exploiting the existence of inverse properties defined with the
construct owl:inverseOf.
2. Properties Domain Range Procedure: this procedure obtains new correspondences
between classes by comparing the domains and ranges of the initial properties. Not
only the first-level domain and range classes are evaluated but the procedure
continues until reaching the higher levels of the hierarchy.
3. Classes Properties Procedure: this procedure allows the retrieval of both new
correspondences between classes and properties. This procedure recursively
identifies the similar properties existing among the class correspondences, and then
assesses the existence of other classes belonging to the domain or range of this
properties that could be a new correspondence.
4. Classes Family Procedure: this procedure retrieves new correspondences between
classes. It exploits the familiar relations of the classes. For each pair of them, its
superclasses, subclasses, and sibling classes are evaluated to determine the existence
of new possible matches.
These procedures are iteratively applied until no new correspondences are
discovered. Once these procedures have finished all the correspondences that have
been discovered are filtered to produce the final output of the algorithm. To do so, the
tagging is very important as it allows the identification of the different procedures and
sub-procedures. It is based on the idea that the different procedures exploit different
structural features of the ontologies and hence the likelihood that the obtained results
are good is not the same for all of them.
To evaluate the performance of the algorithm we intended to use ontologies from
the smart city domain. However, the amount of ontologies in this area proved not to be
enough to allow an accurate evaluation. Hence we have used the testbed provided by
the Ontology Alignment Evaluation Initiative 2013 [19] (OAEI-13) which provides
different series of tests to evaluate the performance of a matching algorithm. This is
usually done by using the standard information retrieval metrics of precision, recall
and f-measure [4].
59
2nd International Workshop on Ontologies and Information Systems
─ Precision: measures the ratio of correct correspondences over the total number of
returned ones. It reflects the degree of correctness of an algorithm.
#_
= ݊݅ݏ݅ܿ݁ݎ (1)
# _
─ Recall: measures the ratio of correct correspondences over the number of
expected ones. It reflects, the degree of completeness of an algorithm.
= ݈݈ܽܿ݁ݎ
#_
(2)
# _
─ F-measure: is a measure introduced to compare the systems with just one value
since it is highly likely that the system with a higher recall may have a lower
precision and vice versa.
݂−
∗
݉݁ܽ = ݁ݎݑݏ (3)
!"∗ # !∗
These measures were used to evaluate the performance of our algorithm. Among the
range of tests at the OAEI-13 we have tested our algorithm with several of them,
although for the scope of this paper we will be focussing on the conference track which
aims at finding alignments within a collection of ontologies from the domain of
conference organization. The results obtained by our algorithm for each pair of input
ontologies are compared with a reference alignment also obtained from the OAEI-13
website. In table 1 we include the average results obtained for this task.
Table 3. Average values obtained in the conference track
Precision Recall F-measure
0.86 0.57 0.67
In the smart cities domain, there is a series of ontologies that describe the resources
and services that are available for the agents. If an agent needs a certain resource or
service, it will need to match its ontology to the appropriate one in the smart city.
Depending on where the service or resource is deployed, the agent may need to match
its ontology to a part of the ontology that describes the smart city itself, usually when
the agent needs access to a resource, or to another agent’s ontology, usually when the
agent needs a service that is offered by the other one. This situation is depicted in
figure 2. In any case, this process will output an alignment between both ontologies. If
several agents need to access the same resource or service, the process will be repeated
several times.
Our intuition is that if a new agent arrives in the smart city and is willing to use a
service or resource, the alignments previously obtained from other agents may help in
tuning the alignment process for this new agent and therefore they may be used to
enhance the results produced by the algorithm.
This led us to delving into alignment reuse techniques [20] which in spite of not
being a particularly used matching technique [4], it was precisely the one that better
60
2nd International Workshop on Ontologies and Information Systems
met the our requirements. This technique is grounded on the idea that when describing
an application domain the ontologies to be matched are similar to already matched
ones and hence this knowledge may be reused. This idea was implemented in the
COMA [21] and COMA++[22] systems which are two of the most well-known ones
and that have been continuously evolving since 2002 to include new matchers and
features.
Fig. 2. Fig. 3. Smart City
To asses the viability of integrating alignment reuse as part of the ontology
matching proposal for smart cities, we have used the ontologies of the conference
track. The procedure followed to do so is to feed the algorithm with some intermediate
alignments that are then used to identify binding points between the ontologies to
match.
Consider the following example, let us suppose that there are three different
ontologies, , and , and that we need to match ontology to ontology . If we also
have available the alignments between and ( __) and, and
( __), then it is possible to identify a path that, using these intermediate
alignments, may link entities in to entities in . We refer to this as a ring between
and through , and it is graphically represented in figure 3.
Tables from 2 to 6 show the results of testing this approach with the ontologies of
the conference track. From the ontologies available at this track in the following
examples we have used the following ones: cmt, conference, confOf, edas, ekaw and
sigkdd.
Table 2 shows the results obtained by directly matching the conference ontology to
the confOf ontology. These values are included to provide a baseline to compare the
results obtained when using rings.
61
2nd International Workshop on Ontologies and Information Systems
Fig. 3. Fig. 4. Ring
Table 4. Results obtained without using any ring
conference - confOf
Precision 0.9
Recall 0.6
F-measure 0.72
Tables 3 and 4 show two different sets of results obtained when using an additional
ontology as ring. Table 3 contains the set of results obtained when matching
conference to confOf, using as additional input the alignments output when matching
conference to edas and edas to confOf. Table 4 presents the set of results obtained
using ekaw. As we can observe, in any case, the values obtained are better than those
in table 2. However, the improvement using edas was more noticeable in the recall,
while the improvement using ekaw was in the precision.
Table 5. Results obtained using edas for the ring
conference - (edas) - confOf
Precision 0.86
Recall 0.80
F-measure 0.83
Table 6. Results obtained using ekaw for the ring
conference - (ekaw) - confOf
Precision 1.00
Recall 0.73
F-measure 0.84
When using edas for the ring, 10 different paths from conference to confOf were
detected which allowed the identification of 5 new correspondences. Using ekaw, just
8 different paths were identified which added 2 new correspondences that were not
detected when directly running the matching process. We considered then a combined
62
2nd International Workshop on Ontologies and Information Systems
approach using both edas and ekaw at a time, seeking to obtain results with the
precision enhancement provided by ekaw and the recall enhancement provided by
edas. The results obtained are shown in table 5.
Using a multiple ring the amount of identified paths rises to 13. The results obtained
with this approach show that precision is not as high as when just using the single ring
with ekaw as there is an extra incorrect correspondence that is added to the final
output. However, for recall and F-measure, the values obtained are remarkably higher
that those obtained with single rings and when compared to the baseline results we
observe an improvement of 10.95% for precision, 62.26% for recall and 35.48% for
f-measure.
Table 7. Results obtained with edas and ekaw used at the same time
conference - (edas &ekaw) - confOf
Precision 0.92
Recall 0.86
F-measure 0.89
In spite of being a positive outcome, the results vary when using the alignments
with other ontologies as rings. Table 6 shows the results obtained when considering the
alignments with cmt and sigkdd for the rings. The results in this table also show an
improvement compared to the baseline in table 2, although they are not as remarkable
as those in table 5.
Table 8. Results obtained with cmt and sigkdd at the same time
conference - (cmt & sigkdd) - confOf
Precision 0.90
Recall 0.66
F-measure 0.76
Other tests run using more alignments showed no improvement compared with
using just two as in the examples presented previously. However the testbed that we
used is not large enough to entirely dismiss that possibility. An issue that we have
identified is that even when using the reference alignments provided by the OAEI,
which are the golden standard used to compare the results of any algorithm, there were
some paths that we identified, that led to selecting as binding points pairs of entities
that then were not considered as a valid correspondence in the reference alignment.
4 Conclusions & Future Work
In this paper we have introduced the smart cities domain and underline some
communication and interaction problems that are highly likely to show up in this kind
of development. We have also described the foundations of the ontology matching
based approach that we propose to tackle such problems in the smart cities, and the
ontology matching algorithm that we have defined to address this problem. We have
63
2nd International Workshop on Ontologies and Information Systems
described the measures of precision, recall and f-measure used to evaluate this type of
algorithm, and the results obtained when doing so.
We have then looked at some alternatives to refine the alignments obtained by our
algorithm and therefore improve such results. Among the different techniques we have
focused on alignment reuse as it was particularly indicated for scenarios like ours. We
have used previously existing alignments, rings, to identify paths of between the
ontologies to match and therefore enhance the results obtained. We have tested this
approach using the ontologies available from the conference track of the Ontology
Alignment Evaluation Initiative 2013. And we have verified the viability and validity
of the approach.
In spite of these good results that account for the viability of our approach, there are
some issues that need to be addressed in order to obtain the best results possible and
hence to improve the usability of the smart cities. There is, for instance, the need to test
our proposal using ontologies taken from the real domain where it will be deployed,
the smart cities. Additionally, as we introduced in section 3, there seems to be a direct
relation between the rings chosen and the goodness of the results obtained, so we aim
at focusing on determining the features that make a ring better than other. It is also
necessary to explore techniques that will allow the alignment reuse in the real
environment, so we are turning to some techniques such as alignment storing and
sharing and alignment annotation.
References
[1]Katole, B., Sivapala, M., V, S.: Principle elements and framework of internet of things.
International Journal Of Engineering And Science 3(5) (07 2013)
[2]Schaffers, H., Komninos, N., Pallot, M., Trousse, B., Nilsson, M., Oliveira, A.: Smart cities
and the future internet: Towards cooperation frameworks for open innovation. In Domingue,
J., Galis, A., Gavras, A., Zahariadis, T., Lambert, D., Cleary, F., Daras, P., Krco, S., Müller,
H., Li, M.S., Schaffers, H., Lotz, V., Alvarez, F., Stiller, B., Karnouskos, S., Avessta, S.,
Nilsson, M., eds.: Future Internet Assembly. Volume 6656 of Lecture Notes in Computer
Science., Springer (2011) 431–446
[3]Chin, J.S.Y., Callaghan, V., Clarke, G., Hagras, H., Colley, M.: End-User Programming in
Pervasive Computing Environments. In Yang, L.T., Ma, J., Takizawa, M., Shih, T.K., eds.:
PSC, CSREA Press (2005) 187–192
[4]Euzenat, J., Shvaiko, P.: Ontology matching. 2nd edn. Springer-Verlag, Heidelberg (DE)
(2013)
[5]Uceda-Sosa, R., Srivastava, B., Schloss, B.: Using Ontologies to make Smart Cities Smarter.
IBM Research. (06 2012)
[6]:SOFIA2. http://scfront.cloudapp.net/ (5 2014)
[7]Wiesman, F., Roos, N., Vogt, P.: Automatic ontology mapping for agent communication. In:
Proceedings of the first international joint conference on Autonomous agents and multiagent
systems: part 2, ACM (2002) 563–564
[8]Shvaiko, P., Euzenat, J.: Ontology matching: state of the art and future challenges. IEEE
Transactions on Knowledge and Software Engineering (2012)
[9]Gal, A., Shvaiko, P.: Advances in ontology matching. Lecture Notes in Computer Science
(including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in
Bioinformatics) 4891 LNCS (2008) 176–198
64
2nd International Workshop on Ontologies and Information Systems
[10]Cohen, W.W., Ravikumar, P.D., Fienberg, S.E.: A Comparison of String Distance Metrics
for Name-Matching Tasks. In: IIWeb. (2003) 73–78
[11]Duman, H., Hagras, H., Callaghan, V.: Intelligent association exploration and exploitation
of fuzzy agents in ambient intelligent environments. Journal of Uncertain Systems 2(2)
(2008) 133–143
[12]Rahm, E., Bernstein, P.A.: A Survey of Approaches to Automatic Schema Matching. Very
Large Data Base (VLDB) Journal 10(4) (2001)
[13]Jiménez-Ruiz, E., Grau, B.C. In: Logmap: Logic-based and scalable ontology matching.
Springer (2011) 273–288
[14]Hanif, M.S., Aono, M.: Anchor-flood: Results for oaei 2009. In Shvaiko, P., Euzenat, J.,
Giunchiglia, F., Stuckenschmidt, H., Noy, N.F., Rosenthal, A., eds.: OM. Volume 551 of
CEUR Workshop Proceedings., CEUR-WS.org (2008)
[15]Noy, N.F., Musen, M.A.: Anchor-prompt: Using non-local context for semantic matching.
In: Workshop on Ontologies and Information Sharing at the Seventeenth International Joint
Conference on Artificial Intelligence (IJCAI-2001), Seattle, WA (2001)
[16]Le, B.T., Dieng-Kuntz, R., Gandon, F.: On Ontology Matching Problems (for building a
corporate Semantic Web in a multi-communities organization). In: Proc. of the Sixth
International Conference on Enterprise Information Systems - Porto - Portugal, Kluwer (14-
17 April 2004)
[17]Miller, G.A.: WordNet: a lexical database for English. Communications of the ACM
38(11) (1995) 39–41
[18]World Wide Web Consortium. W3C: OWL: Web Ontology Language (2014)
[19]Euzenat, J., Meilicke, C., Stuckenschmidt, H., Shvaiko, P., dos Santos, C.T.: Ontology
Alignment Evaluation Initiative: Six Years of Experience. Journal on Data Semantics -
Springer (2011) 158 – 192
[20]Rahm, E., Bernstein, P.A.: A survey of approaches to automatic schema matching. The
VLDB Journal 10(4) (2001) 334–350
[21]Do, H., Rahm, E.: COMA - a system for flexible combination of schema matching
approaches. In: Proceedings of the 28th VLDB Conference, Hong Kong, China (2002)
[22]Aumueller, D., Do, H.H., Massmann, S., Rahm, E.: Schema and ontology matching with
COMA++. In: Proc. ACM SIGMOD Conference. (2005)
[23]Naphade, M.R., Banavar, G., Harrison, C., Paraszczak, J., Morris, R.: Smarter Cities and
Their Innovation Challenges. IEEE Computer 44(6) (2011) 32–39
[24]Uceda-Sosa, R., Srivastava, B., Schloss, R.J.: Building a Highly Consumable Semantic
Model for Smarter Cities. In: Proceedings of the AI for an Intelligent Planet. AIIP ’11, New
York, NY, USA, ACM (2011) 3:1–3:8
[25]Villanueva, F.J., Santofimia, M.J., Villa, D., Barba, J., López, J.C.: Civitas: The Smart City
Middleware, from Sensors to Big Data. In Barolli, L., You, I., Xhafa, F., Leu, F.Y., Chen,
H.C., eds.: IMIS, IEEE (2013) 445–450
[26]Fazio, M., Paone, M., Puliafito, A., Villari, M.: Heterogeneous Sensors Become
Homogeneous Things in Smart Cities. In You, I., Barolli, L., Gentile, A., Jeong, H.D.J.,
Ogiela, M.R., Xhafa, F., eds.: IMIS, IEEE (2012) 775–780
65