CLUSTERING-BASED SERVICE SELECTION FOR DYNAMIC SERVICE COMPOSITION

International Journal of Web & Semantic Technology (IJWesT) Vol.8, No.2, April 2017
DOI : 10.5121/ijwest.2017.8201 1
CLUSTERING-BASED SERVICE SELECTION FOR
DYNAMIC SERVICE COMPOSITION
Aram AlSedrani and AmeurTouir
College of Computer and Information Sciences, Department of Computer Science King
Saud University, Saudi Arabia
ABSTRACT
The increase in the number of available web services led to the increase in the similarity of services
functionality offered by different providers each with different QoS parameters. Therefore, in web service
composition, the selection of the optimal service to satisfy the QoS values required by user is one of the
significant requirements. Moreover, the dynamic nature of web services adds more challenges to obtain the
accuracy of the selection process. Most of the existing service composition approaches deal with services
changes during composition execution, causing a re-planning or re-selection that affecst the service
composition performance. In this paper, we introduce the clustering-based service selection model that
outperforms the existing ones. The proposed model has the ability to detect and recover the changes in
service repository by monitoring the composition process from a global point of view. The approach is a
two-levels-based web service clustering. The proposed model encompasses a clustering process, a planning
process, a selection process and a recovery process.
KEYWORDS
Web services, service clustering, service composition, service selection, self-healing composition
1. INTRODUCTION
Web service composition involves the integration of several existing web services to provide
more complex and powerful service. The goal of service composition is to reuse existing web
services and composing them into a process. However, the process of composition is considered
as a high complex task due to many reasons. According to (Rao & Su 2004) the complexity of
service composition raises because of three main reasons. First, the huge increase on the number
of web services available over the internet, whichcauses the increase of the service repositories
available for searching. Moreover, web services is in continuously changing and updating which
requires a dynamic composition at runtime causing intensive computations. Another reason when
different providers develop web services in different models making the mapping between
services in the compositions a difficult task.
Service selection is about choosing the most appropriate service for each task in the composition
in order to satisfy user requirement.The input of service selection problem is the set of candidate
services for various tasks involved in the composition plan. A single candidate services set
consists of services providing the same functionality offered by different providers through
service Quality of Service (QoS) profile. Moreover, the same provider may offer several services
that have the same functionality with different QoS to obtain satisfaction of a large panoply of
users(Moghaddam & Davis 2014).
Nowadays, the composition environment becomes more dynamic due to the increase in the
number of web services that are frequently changing. Therefore, the need for self-adapted

2
composition methods that acts according to environment changes is advocated. Some existing
composition approaches consider more or less the dynamic environment context while executing
services. This is performed by monitoring the changes that occur to services, and considering the
environment as static in early phases as in (Friedrich et al. 2010) and (Wiesner et al. 2009).
A limited number of researches highlight the importance of monitoring the changes on services
during the selection phase. Batakat(Barakat et al. 2012) proposes an algorithm that reacts to
changes that occurred in the participating web services during the selection phase. The algorithm
modeled the problem as a graph, where each node stores the optimal path from the started node.
When a change occurs, the algorithm re-tracks the stored information in each node in the path and
reselects a new service accordingly. The proposed algorithm runs in a dynamic environment.
However, when the number of nodes increases the complexity and the storage capacity increases
considerably.
The limited number of researches in this area indicates that the service selection in dynamic
environment is a new research direction, and it is not mature enough to satisfy composition
process requirements. The need for new techniques is highly recommended, especially with the
fast growth of the web.
In this paper, we propose the self-healing model for web service selection that has the ability to
recover from service failure. The self-healing model involves three main processes: the web
service clustering process, the service selection process, and the failure recovery process. The
service clustering process involves two levels of clustering: first, classify the similar web services
based on their functionality to “task”, second, generate composite tasks and group them into more
abstract representation called “job”. By two levels services clustering, we aim at three goals:
First, minimizing the selection problem search space within jobs rather than services. Second, the
ability of fast recovery, in case of service failure, by substitutes the faulty service from the
corresponding job and task clusters. Third, isolate the faulty web service in its job boundary and
not affecting the rest of the well composed service. The selection in the proposed model is
monitored through its entire process, where the failure recovery process is triggered whenever an
environment change is occurred.
2. THE SELF-HEALING MODEL FOR WEB SERVICE SELECTION:
We define the Web Service Selection Problem (WSSP) as finding the most suitable web services
from a set of candidate web services that match the QoS attributes asked by user. The adopted
self-healing model embodies the service clustering process, the service selection process, and the
failure recovery process. The service clustering classifies the web services with similar
precondition and post condition to tasks. Thus, each task includes a set of atomic web service. As
a second level, the service clustering classifies similar tasks into more abstract representation
named jobs. A job represents a set of simple or composite tasks that have the same precondition
and post condition. A composite task is composed of a sequence of tasks. The selection process is
responsible of selecting the best candidate web service (low level) related to each job (high level)
involved in the planning phase based on its QoS values. However, if any contingency occur
during the selection process, the failure recovery process is triggered to react and solve the
emerging problems.
As it was stated in the formulation of the WSSP, we used two types of service clustering in order
to arrange the services prior the selection process: the task clustering and the job clustering. Each
(J), includes multiple candidate tasks either atomic or composite. The principle of web service
selection algorithm is to substitute each job from the composition plan by a corresponding web
service via the clustered tasks in order to achieve the requested service (RS) and satisfy the QoS

3
requirements. The input of the selection algorithm is the QoS Constraint Plan (QCP) that consists
of set of connected jobs, and the solution is the Execution Plan (EP) that consists of set of
connected web services. The QoS attributes that are going to be addressed by our model are the
Cost (C) and the Execution Time (ET). The Cost (C) is the price that a user has to pay to the
service provider for theinvocation of a service (). The Execution Time(ET)is the time measured
from invoking () until theresult is received(Rajeswari et al. 2014). Moreover, the failure
recovery algorithm is triggered in case of web service failure. It aims at collaborating with the
selection process to recover the failures by substituting the faulty web service with a valid service
from the same job cluster. In the sequel, we will elaborate the self-healing model processes in
more detail.
2.1. DEFINITION AND SPECIFICATION:
In the sequel, we use the following definitions and notations:
- A web service directory WS is the set of services of a specific domain that exist in a service
repository. We assume that each web service  WS is atomic that is not composed of other web
services. Each  WS has two values: a cost Cthat specifies the cost of executing , the
execution time ET that denotes the execution time of.
- A web service (, , ) is an atomic web service, where  is the precondition,  is the effect
and  is its QoS.
- A task (, , , , WS, ) is an abstract representation of all web services with similar
functionality. That is the set of the web services which have the same precondition  and the
same effect . The QoS of ,  =   ,  is the number of web service in , WS is the
set of equivalent web services, and is the candidate web service to be selected with quality . T
is the set of all tasks.
- A job J(, , q, ,Tj, ) is the cluster of the tasks that have the same precondition , the same
effect . The QoS of J, q =  
, and finally,  is the number of tasks in J, Tj is the set of
tasks, and is the candidate task to be selected with quality q.
-A QoS Constraint Plan is a directed acyclic graph denoted as QCP [J(, ), E, Q] where: J(,
)={j1 … jn/ji is a job i = 1, …, n} is the set of vertices representing the jobs in the graph where
the precondition of j1=  and the effect of jn =. E J X J is the set of edges. Q is the global QoS
for the plan Q= . We assume that the jobs flow sequentially within QCP starting from
j1 until jn.
-We define the Web Service Selection Problem (WSSP) as follow: given a QoS Constraint Plan
QCP, the WSSP is to find the optimal web service  for each j ϵ J that has the best quality by
satisfying the following conditions:
C() = min { C(i) , i ϵ j } -1-
ET() = min { ET(i) , iϵ j } -2-
-The Execution plan (EP) is defined as the solution of WSSP. It is denoted as EP={ (j1, 1), (j2,
2), ……, (jn, n)} where:

4
a. {j1, j2, ….. , jn } ϵQCP.
b. (ji, i) ϵ EP (i ϵ ji)and (i ϵ i ).
2.2. TASK AND JOB CLUSTERING:
As it was stated, the repository is made up of two-level of abstractions. The jobs: where each
jobincludes a set of tasks having the same preconditions and effects. The tasks: where each task is
a generic abstraction representing a set of serviceswith similarpreconditions and effects. The
clustering is a continuous process thatupdates the tasks and jobs repositories. The clustering is
built from the ground-up.
Each newly discovered web service () isinsertedinto the Web Service repository WS and the
update is propagated accordingly to the Task repository and the Job repository. The update
propagation takes into account the definition and theconstraints (precondition and effect) of the
discovered web service.
The first step in the clustering process is to update the task repository. For each web service (,
, ), the precondition () and effect () constraints are extracted from the service description
profile. Then, match the constraints with the available tasks in task repository and insert  into its
belonging task. If no task matches the service constraints a new task  is added to the task
repository and assigned precondition ., quality .and effect .from profile. For each
inserting , the QoS  is revised as to keep .with the minimum of all web services within the
task cluster.
Several QoS attributes are considered for a single web service. In such problems a utility function
is used to evaluate the multi-attributes quality of a set of alternative web services. For this
purpose, we use the multiple attribute decision making, in particular the Simple Additive
weighting (SAW) method (Hwang & Yoon 1981) is used. The evaluation score is calculated
based on multiplying the scaled value for the alternative web services with the importance weight
of the quality attribute, then adds the product for all attributes. The SAW uses the following
formula to evaluate the alternatives.
  -3-
Where  is the evaluation score of the web service , is the weight of the attribute a, and
 is the score of  with respect to attribute a. The maximum the value of  means the
minimum the quality attribute values, thus the better quality. The value of  is calculated for
each web service within the same task cluster  for each quality attribute based on the following
formula:


-4-
For task clustering process, we evaluate and rank the quality of each web service within the same
task cluster based on the values of execution time ET() and cost C(). These quality values are
extracted from the OWL-S description file for each web service . Knowing that, OWL-S is Web
Ontology Language for describing Semantic Web Services.
On the other hand, creating the Job repository is made up in two times. First, for each task , a job
J is created. The job J takes the precondition ., the quality ., the effect ., and the singleton
{} in its task set. As two simple tasks cannot have the same precondition and effect, a job J may

5
have at most one simple task. Composite tasks may be added later based on the circumstances.
Each job will be updated by generating composite tasks with the same constraints. That is, if a
task  fails to achieve the request, a new composite task is prepared to substitute the failed task.
The values of the job quality q, the job length  and the candidate task are updated
accordingly.
The quality of the job j.q is calculated using SAW method. Based on the quality .andlength
.of the task , we compute the SAW of each task. Then we compute the SAW of the job. The
job quality j.q will equal the Maximum of all SAW values that reflect the minimum of the
execution time and the cost resulting with the best available quality.
The quality .and the length .for a composite tasks in each job cluster are the aggregation of
these values in each task. The aggregation values for the Cost (C) and Execution Time (ET) for a
composite task are the summation of their respective values in each task involved as shown in
equation (5, 6).
 
 
 
 
2.3. CLUSTERING-BASED WEB SERVICE SELECTION:
As defined in section 2.1, the WSSP aims at finding the optimal web service to compensate each
job in QCP plan. This optimality is achieved by finding the web services with the best Quality
values. In our model, we utilize the two QoS attributes (Execution time, Cost) to direct the
selection process. In the clustering process, each service is inserted in a task cluster .
Furthermore, the web service with minimum quality is identified in the cluster as the candidate
web service to be chosen (. ). This process is also applied to the job repository for each new
task inserted to any job cluster by identifying (J. ).
By preparing the candidate web service (. ) for each task, the Clustering-based Selection of
web service as described in algorithm 1 is based on substituting each job in the QCP with its
candidate task ; then, substitute each task with its candidate web service . The output of the
selection process is the execution plan (EP) as defined in section 2.1.
Algorithm 1:Clustering-based Selection
1: Procedure Clustering-based Selection (CP)
2:Foreach ϵ CPDo
3: Select Where .=

(SAW)
4: If.getLength() =1Then
5: Select ϵ Where .=
 ϵ 
(SAW)
6: EP.Add (, )
7: EndIf
8: If.getLength() > 1 Then

6
9: suppose c :composite empty task
10: Foreach( ) Do
11: Select ϵ  Where .=
 ϵ 
(SAW)
12: c.Append(’)
13:EndForeach
14: EP.Add (c, )
15: EndIf
16: EndForeach
17: Return EP
2.4 SELF-HEALING SELECTION ALGORITHM
The self-healing selection algorithm has the ability to recover any changes or faults that occur
during selection phase of service composition process. It aims at substituting the faulty web
service with similar web service from the same job/task cluster. Algorithm 2 describes the self-
healing process; In case of web service failure, the algorithm test if the failed service belongs to
an atomic task, if so, the failed service is replaced with an alternative one from that task. In the
other hand, if the failed service is within a composite task, then replace it with the next best
alternative web services based on its quality, and then recalculate the composite task quality. The
next step is to determine if the composite task still has the minimum quality comparing with the
tasks in its job cluster, then it will be substituted by job in EP. Else select an alternative task from
the same job with the minimum task quality.
Algorithm 2:Self-healing Selection
1: ProcedureSelf-healing Selection(EP, )
2: <- .getTaskOwner()
3: foundAlternative<- False .
4: If.getLength() > 1 Then
5: Re-Calculate .
6: Select ’ ϵ  with ’.=
 
(SAW)
7: EP.Substitute(;’)
8: foundAlternative<- True
9: EndIf
10: If.getLength() =1 Then
11: j <- .getJobOwner()
12: Re-Calculate j.q
13: Select ’ ϵ with ’.=

(SAW)
14: suppose c :composite empty task
15: Foreach( k ϵ ’ ) Do
16: Select ’ ϵ  with ’.=
 
(SAW)
17: c.Append(’)
18: EndForeach
19: EP.Substitute(;c)
20: foundAlternative<- True
21: EndIf
22: ReturnfoundAlternative

7
Example 1:consider the following clusters available in the task repository(table 1) with their QoS
values. Each web service has two quality values, the first is the execution time ET measured in
msand the second is the cost C measured in $:
Table 1: Web Service Repository Example
For each web service, we apply the SAW formula to calculate the normalized quality values. By
assigning, for example, the weights of the criteria as W (ET)= 0.5 and W (C)= 0.3 meaning that
the ET is more important than C. However the weight of a quality criterion could be adjusted
according to the user preference. Based on this assumption, we can calculate the quality value for
3 (a,b) as follow:



 


And accordingly, the quality values for the web services in each task cluster in the task repository
will be as shown in Table 2

8
Table 2: Task Repository Example Including QoS
Accordingly, The quality values for the tasks in the job repository is computed using eq. (5, 6),
knowing that a task  has two attributes as ranking criteria ., and .. For example to calculate
the quality of (1 5) in J2 (Table 3), we will apply eq. (5, 6), taking W(.)= 0.5 and W(.) =
0.3:
 
 
   
 
   
 
   
   
Since   
=  
 
= 0.61 + 0.5 = 1.11
And    =   +   = 6 + 5 = 11
Then   
 
Table 3 shows the job repository results after applying SAW formula.

9
Table 3: Job Repository Example
Example 2:
Consider the followingCPthat represents a plan with a precondition (a) and an effect (e) generated
according to the jobs in previous table:
Figure 1: Quality Constraint Plan Example
From the job and task repositories we can retrieve the following information:
Table 4: Service Selection Example
Job
J4 4 3
J3 3 3
J2 2 4

10
According to table 4, the Execution Plan (EP) will be:
To illustrate the selection algorithm in case of service failure, we consider several situations; each
example is independent of the other:
Situation 1:Suppose we have an event occurring in service repository that deletes J4.2.4. In this
case, the deleted web service does not affect the selection process and nothing changed since 4
is not the candidate service in task cluster 2.
Situation 2:Suppose we have an event occurring in service repository that deletes the web
services J2.2.1 and J2.2.4. In this case, the selection is affected and the algorithm will first
recalculate the SAW values for the remaining web services in J2.2.
2 (b, e) {2 (b, e).=0.3 3 (b, e). = 0.5} 2. = 2
Then, recalculate the SAW values for the tasks in J2 and select the web service with maximum
quality value.
J2 {2.=0.8 (1 5).= 0.3 (6 4 3).=0}
The web service with maximum quality value= J2.2.3. In this situation the Execution plan (EP)
will result with:
2.5. TIME AND SPACE COMPLEXITY:
Lemma: For a web service failure within a job J in a given CP, the time complexity of the Self-
healing Selection algorithm depends on the number of web services in the job J.
The recovery process aims at substituting the failed web service with a valid one from the same
job. Therefore, the time needed to recovery is the time to traverse all the web services in the job
to recalculate the quality, thus selecting the optimal service.
The number of services in a job is equal to all the services within the atomic and composite tasks.
Accordingly, the number of services in a single composite task  (that composed of nh simple
tasks) and each simple task (hi) has nhi web services=
Thus, total number of web services in the job (where m is the number of tasks in j) is

11
3. PERFORMANCE EVALUATION:
3.1. DATASET AND EXPERIMENT SETUP:
In our experiment, we utilized several web service description files datasets to test our model,
The OWLS-TC version 4.0 dataset, OPOSSum database for service descriptions [8], and SWS-
TC-1.1 dataset provided by [9]. The OWLS-TC dataset provides 1083 semantic Web services
written in OWL-S 1.1 from nine different domains (education, medical care, food, travel,
communication, economy, weapons, geography and simulation)[10]. The OPOSSum is an open
source project to collect semantic web services descriptions launched by Friedrich Schiller
University in Germany, the data collection include 1129 OWL-S file in different domains. The
SWS-TC-1.1 is a public dataset that contains 241 semantic web services descriptions in OWL-S
also. Moreover, we collect extra web service descriptions from the web in order to raise the
number of services and we came up to 6785 OWL-S services description files. After performing
the job cluster algorithm we came up with 739 job clusters in the job repository and the same
number in task repository.
However, the OWLS-TC dataset does not contain any information of the Quality values of the
web services. Therefore we generate random values of Execution Time and Execution Cost that
are picked from the QWS DATASET 1.0[11]. In our model we used only the Execution time and
Cost quality parameters. Though, the model can be easily generalized to include other QoS
parameters. The parser in our model extracted the main services parameters needed to the
composition process from OWLS-TC dataset such as service name, input, output, preconditions,
effect, quality values, and service description.
In order to evaluate the performance of the clustering-based selection and the self-healing
algorithm we implementedtwoother algorithms that react to changes occurred in dynamic
repository during selection phase.The algorithm provided by Barakat in [5] that stores the optimal
path from the initial to the goal in each node in the plan.The traditional method of re-composition
in case of faults occurs.
3.2. EXPERIMENTS RESULTS:
The experiment involves 6785 OWL-S files in the service repository, and after performing the job
cluster algorithm we came up with 739 job clustersin the job repository. We performed two
experiments in order to measure the performance of the selection and of the recovery in case of
service fail.
3.2.1. Clustering based selection time:
In this experiment, we apply 100 test cases to measure the selection time of our method. For each
test case the input is a composition plan (CP) that consists of jobs and the output is an execution
plan (EP) that involves of the candidate web services. The plans vary in length; the range in our
experiment start with plans consisting of 50jobs to plans containing500 jobs. Figure 2 shows the
selection time resulted from the experiment of the clustering-based method. The time needed to
select web services for each job within a plan is between 0.9 and 2.6 seconds.

12
Figure 2: Clustering-based Selection Time
To highlightthe performance of the developed selection methodwith respect to the existing
methods, we compared the performance of the clustering-based and barakat methods. Figure 3
shows the major enhancement in the selection time by the clustering-based selection. As one can
notice, the execution time in barakat algorithm is much sloweras it requires storing all the paths
from an exact node in the plan until the goal node. However, in our method, preparing in
advance, the candidate services in each task cluster decreases considerably the time complexity of
the selection process as shown in section 2.5.
Figure 3: Selection Time Comparison
3.2.2 Self-healing Recovery Time:
For this experiment, we picked a random set of plans and executed 100 test cases over the three
methods: self-healing, barakat and reselection algorithms. For each test case, we performed the
following:
 A random plan is entered to the selection algorithm.
 Two types of web service faults are simulated:
o The first is the case when the faulty web services are recovered directly from either task
or job repository. In this type, we performed three failed services; one that occurred in the
0.00
0.50
1.00
1.50
2.00
2.50
3.00
100 200 300 400 500
Clustering based selection time
0
5
10
15
20
25
30
35
100 200 300 400 500
TimeinSeconds
Number of Elements
Service Selection Time
Self-healing
Barakat

13
beginning of the plan, another, when a fault occurred in the middle of the plan and a third,
when the fault occurred in the end of the plan.
By applying the input and simulation of several web service faults described previously, we
ended up with 600 test cases to measure the time needed from the self-healing algorithm to
recover a web service fault during selection phase.For each type of fault simulation, we calculated
the average time needed to recover based on the number of services within the plan. The
evaluation procedure is based on the number of the simulated faults. For each composite service
length (50 -500), we applied first 10 faults during selection, then in the next step we increase the
number by 10 until we simulated 60 faults. We will show the performance for the algorithms for
each number of simulated faults (10 to 60).
The result of our experiment is shown in figure 5.It compares the three methods together.Figure 5
shows that the self-healing algorithm outperforms the other methods in the recovery time in
selecting a substitute service for a faulty one. By preparing the tasks clusters and jobs clusters
prior the selection time, the recovery time will cost the time needed to retrieve the job cluster
where the fault occur and pick the service within that cluster that has the minimum quality values.
Figure 4:Recovery Time Comparison
4. CONCLUSION:
In this paper, we proposed a web service selection algorithm and a self-healing algorithm for
service composition that react to changes in dynamic repository. The algorithm consists of three
phases: first, continuous clustering of the similar web services to tasks and similar tasks to jobs.
Second, a selection phase that chooses the most candidate web service from the task cluster to
replace it in the plan. Third, a self-healing selection algorithm that reacts to changes in the web
service repository by substituting the faulty service with similar one within the same cluster. An
experiment is performed to test the selecting and recovering time for the proposed algorithm.

14
Moreover, a comparison with similar studies is done that shows an improvement in recovery time
resulted from the proposed clustering algorithm.
REFERENCES:
[1] J. Rao and X. Su, “A Survey of Automated Web Service Composition Methods,” in Proceedings of
the First international conference on Semantic Web Services and Web Process Composition, 2004,
vol. 3387, pp. 43–54.
[2] M. Moghaddam and J. G. Davis, “Service Selection in Web Service Composition: A Comparative
Review of Existing Approaches,” in Web Services Foundations, First., A. Bouguettaya, Q. Z. Sheng,
and F. Daniel, Eds. New York: Springer, 2014, pp. 321–346.
[3] G. Friedrich, M. Fugini, E. Mussi, B. Pernici, and G. Tagni, “Exception Handling for Repair in
Service-Based Processes,” in IEEE TRANSACTIONS ON SOFTWARE ENGINEERING, 2010, vol.
36, no. 2, pp. 198–216.
[4] K. Wiesner, R. Vacul, M. Kollingbaum, and K. Sycara, “Recovery Mechanisms for Semantic Web
Services,” in International Conference on Distributed applications and interoperable systems (DAIS),
2009, pp. 100–105.
[5] L. Barakat, S. Miles, and M. Luck, “Reactive Service Selection in Dynamic Service Environments,”
in First European Conference in Service-Oriented and Cloud Computing, 2012, pp. 17–31.
[6] M. Rajeswari, G. Sambasivam, N. Balaji, M. S. Saleem Basha, T. Vengattaraman, and P.
Dhavachelvan, “Appraisal and analysis on various web service composition approaches based on QoS
factors,” J. King Saud Univ. - Comput. Inf. Sci., vol. 26, no. 1, pp. 143–152, Jan. 2014.
[7] C.-L. Hwang and K. Yoon, Multiple Attribute Decision Making: Methods and Applications A State-
of-the- Art and Survey, 1st ed. Munich: Springer-Verlag Berlin Heidelberg, 1981.
[8] Opossum, “OPOSSum, an Online Portal for Semantic Services,” FUSION group, Friedrich Schiller
University Jena, Germany, 2010. [Online]. Available: http://fusion.cs.uni-jena.de/opossum/.
[Accessed: 12-Feb-2017].
[9] Y. Ganjisaffar and H. Saboohi, “SWS-TC 1.1,” SemWeb Central, 2006. [Online]. Available:
http://projects.semwebcentral.org/projects/sws-tc/. [Accessed: 12-Feb-2017].
[10] M. Klusch and P. Kapahnke, “OWLS-TC,” SemWebCentral, 2010. [Online]. Available:
http://projects.semwebcentral.org. [Accessed: 01-Jan-2016].
[11] E. Al-Masri and Q. H. Mahmoud, “QoS-based Discovery and Ranking of Web Services,” in IEEE
16th International Conference on Computer Communications and Networks (ICCCN), 2007, pp. 529–
534.
AUTHORS
Prof. AmeurTouir, teaching in King Saud University in Riyadh KSU, he got his B.Sc. in
computer science from Saint Etienne University in 1985. And his M.Sc. degreefrom Pierre &
Marie Curie (Paris VI) University in Artificial Intelligence 1988.Ph.D. in Computer Science
1993 from Ecole Nationale Superieure des Telecommunications de Paris. Prof.Touirresearch
interests is in Databases, spatial access methods, spatial query processing, spatio-temporal,
object-oriented, design pattern.
Aram AlSedrani is a PHD student in King Saud University KSU in Riyadh, KSA. She got her B.Sc.
degree from KSU in computer science in 1998, and got her M.Sc. degreefrom KSU in 2009. Aram
research interest in software engineering, service oriented computing

CLUSTERING-BASED SERVICE SELECTION FOR DYNAMIC SERVICE COMPOSITION

More Related Content

What's hot (12)

Similar to CLUSTERING-BASED SERVICE SELECTION FOR DYNAMIC SERVICE COMPOSITION (20)

Recently uploaded (20)

CLUSTERING-BASED SERVICE SELECTION FOR DYNAMIC SERVICE COMPOSITION