Balancing between Local and Global Optimization of Web Services
Composition by a Fuzzy Transactional-aware Approach
Fatma Rhimi
, Saloua Ben Yahia
and Samir Ben Ahmed
Université de Carthage, INSAT, LISI Laboratory, Tunis, Tunisia
ESPRIT, School of Engineering, Tunis, Tunisia
Keywords: Fuzzy Logic, Optimization, Quality of Service, Skyline, Transactional Selection.
Abstract: The tremendous growth in the amount of available web services due to the proliferation of paradigms such
as Big Data and Cloud Computing has raised many challenges in service computing. When there are
multiple web services that offer the same functionalities, we need to select the best one according to its non-
functional criteria (e.g. response time, price, reliability) while guaranteeing a global optimization. Many
approaches have been introduced to tackle this problem. However, most of them neglected users
preferences, which can be very vague and imprecise, in the selection process. Besides, transactional
properties that can insure a reliable achievement of the composition are rarely considered. This paper
suggests a solution to this challenge by modelling users uncertain preferences with fuzzy sets. We then
compute the set of Skyline services which are the best candidates in the search space with fuzzy dominance
relationship and fuzzy similarity measures. Finally we inject transactional properties in order to guarantee a
global optimization with a successful achievement of the composition process. Experimental evaluation
demonstrates the effectiveness of the proposed concept and the efficiency of our implementation.
Web Services are growing in popularity as an
efficient solution to enhance the interoperability for
machine-to-machine interaction among different
applications and different platforms. This is why
business structures are moving today towards the
service-oriented architecture as web services seem to
be the best solution to allow the exchanges between
them. Service composition is a process that
combines multiple atomic web services in order to
create value-added web services. Hence, it is arising
as an effective solution to deliver customized
services to the different users.
Quality-of-Service (QoS) is widely employed to
represent the non-functional characteristics of web
services and has been considered as the key factor in
service selection. QoS is defined as a set of
properties including response time, throughput,
availability, reputation, etc. However, today with the
prevalence of paradigms such as Big Data, Cloud
Computing and XAAS (everything as a service), the
number of available web services had exploded.
This is why it has become difficult to choose the
best candidates that would insure an optimal
composition. By optimal composition we mean a
composition that corresponds the most to the
functional and non-functional criteria provided by
the user.
Skyline is a technique that comes as a solution
that helps reducing the search space based on a
dominance relationship to preselect the best services
and prune the others. Intuitively, a skyline query
selects the “best” or most “interesting” points with
respect to all dimensions. Thus, it can be very
effective for reducing the number of candidates and
enhancing the optimization.
On another hand, transactional properties of web
services are crucial to insure the coherence of the
composition process. In fact, delivering reliable
service composition is very important for the overall
quality of the service. Invoking distant atomic web
services which interoperate with each other can be
affected by failures, throughput, availability etc.
Hence, insuring a reliable service becomes as
important as delivering a service which meets users
functional and non-functional requirements.
In this study, we argue that users preferences as
well as transactional behavior of services should be
considered in the selection process. Thus, we
Rhimi, F., Yahia, S. and Ahmed, S.
Balancing between Local and Global Optimization of Web Services Composition by a Fuzzy Transactional-aware Approach.
DOI: 10.5220/0005941200750082
In Proceedings of the 11th International Joint Conference on Software Technologies (ICSOFT 2016) - Volume 2: ICSOFT-PT, pages 75-82
ISBN: 978-989-758-194-6
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
suggest an approach that will recommend to the user
the appropriate service according to his preferences
using fuzzy techniques. Then, we will inject
transactional properties in order to insure the
reliability of the delivered service.
1.1 Contributions
This paper aims to present a new approach for web
services selection which balances between local
optimization by selecting services that correspond to
the users preferences, and global optimization by
respecting transactional properties. The first step is
the computation of the set of Skyline services.
Traditional Skyline computation methods rely on
Pareto dominance relationship: a service p
dominates another service q if p is at least as good as
q in all the dimensions and strictly better in at least
one dimension. Such strict dominance relationship
suffers from looseness and does not accord an
importance to the user’s preferences. Besides, users
behavior is subjective, vague and imprecise. We
suggest a two steps approach:
- We will in first place select the services with the
best matching degrees with the user’s
preferences based on fuzzy similarity measures.
- We will inject transactional properties in order to
guarantee a reliable successful composite
Considering all this, our main contributions may be
summarized in the following points:
- We will address the problem of computing
service Skyline with a consideration of user’s
preferences by making use of fuzzy preference
relationships rather than Pareto dominance
- We compute the matching degree between web
services and users requirements using fuzzy
similarity measures.
- We proceed to optimizing the overall
composition using transactional properties of
web services.
- We evaluate the efficiency and the effectiveness
of the proposed method with a theoretical study
and an experimental evaluation.
1.2 Outline
Section II presents the related work of this study .In
section III we present the background of this work
so we can advance the followed approach. Section
IV will describe the different steps of the proposed
approach. In Section V, we present the experimental
evaluation of the approach. Finally, section VI will
conclude the paper.
2.1 Web Services Composition and
Skyline Computation
The problem of QoS-based web service selection
and composition has received a lot of attention
during the last years. Local selection methods using
techniques such as Simple Additive Weighting
(SAW) were conducted to select services that ensure
an optimal composition. However, local selection
could not satisfy global constraints on the
composition as it treats each service class
individually. Zeng et al., (2003) tackled this problem
using a global planning composition based on mixed
integer Programming technique for dynamic and
quality-driven selection. However, the costs of this
approach are exponential in a large space. Linear
programming methods are very effective when the
size of the problem is small, but suffer from poor
scalability due to the exponential time complexity of
the applied search algorithms. In their work, Alrifai
and Risse (2009) proposed a hybrid selection
approach that combines local selection with global
selection by decomposing global constraints into
local constraints in order to find close-to optimal
solutions. Canfora (2005) proposed a genetic
algorithm to the QoS-based composition. Genetic
algorithms are based on the evolution theory and in
opposition to linear programming algorithms, the
input data doesn’t need to be linear. Besides, genetic
algorithms are related to the number of service
classes and not to the number of candidate web
services, so they are more effective in a large space
context. However, linear programming is proved to
be faster than genetic algorithms and is preferred
hence in a small space. Yu and Keiw-Jay (2004)
proposed heuristic algorithms that can come as an
alternative to exact solutions. The authors modelled
the problem as combinatorial problem and proposed
a heuristic Branch and Bound algorithm (WS HEU)
and a heuristic graph model (MCSP-K). The two
algorithms are proved to be more efficient than exact
algorithms. Ardagna and Pernici (2007) tried to
overcome the shortcomings of both local and global
service composition by proposing an approach that
addresses optimization problems under severe QoS
Skyline is a mechanism that interferes as a filter
in the search space that would select only the
ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends
interesting points. Hence, by discarding all the
irrelevant data points, the number of possible
combinations would be dramatically reduced. The
analysis of the Skyline was originally considered as
a mathematical problem. It was then introduced in
the first place in the field of database. Given a set of
points in d-dimensional space, the Skyline is defined
as the subset containing the points which are not
dominated by another point. Paradigms like Block
Nested Loops (BNL) and Divide to Conquer are
among the first attempts to solve the computing of
Skyline. BNL algorithm uses a nested loop to
compare the points. Divide to Conquer algorithm
recursively divides the whole space into small sets
(regions), calculates the Skyline for each region
separately, and merge them in the final Skyline.
BNL has proposed the Sort Filter Skyline (SFS)
algorithm, which adopts a pre-sorting to improve the
efficiency. SFS has also been improved by linear
Elimination Sort Skyline (LESS), which operates on
a small set of best data objects to remove other
objects in the original passage of the external sort.
The index structures such as B-trees have also been
utilized to improve the performance of analyzing the
Skyline. Nearest Neighbour (NN) and Branch and
Bound Skyline (BBS) are two representative
algorithms that can progressively address the
Skyline based on R-tree structure. In recent works,
many researchers focused on computing skyline
services in the context of service composition.
However, the majority of these works relied on
Pareto dominance relationship for this purpose
Alrifai et al., (2010), Chen (2014), Abourezk and
Idrissi (2014). Pareto dominance has the
shortcoming of neglecting the smoothness and
fuzziness of human preferences. Benouaret et al.
(2011) addressed this problem with a fuzzy
dominance relationship. Fuzzy logic was addressed
in the optimization techniques for service
composition in many contributions such as those of
Almulla et al., (2010), Torres et al., (2011), Xuan
and Tsuji (2008) and Wang (2006).
2.2 Transactional-aware Service
Due to the characteristics of web services including
loosely coupled and heterogeneous nature, insuring a
reliable execution of services becomes very
challenging. Liu et al., (2006) proposed rules for
ensuring transactional services in parallel and
sequential compositions. Hadded, Manoeuvrier and
Rukoz proposed in their leading study (2010) an
algorithm combining QoS optimal web services
selection with transactional constraints to generate
transactional composite services. Bhiri et al., (2006)
proposed an algorithm to check the validity of a
composition according to the set of acceptable
termination states of a transactional composite web
2.3 Fuzzy Similarity Measures
Fuzzy sets were introduced by Lotfi Zadeh (1965) as
an extension of the classical notion of sets.
Fuzzy similarity measures are measures used to
compute the degree of similarity between two fuzzy
sets. The concept of similarity is interpreted in
different ways depending on the context. The
interpretation of similarity in everyday language is
“having characteristics in common”. This
interpretation of similarity differs from the one we
use. We define similarity between fuzzy sets as the
degree to which the fuzzy sets are equal. This
definition is related to the concepts represented by
the fuzzy sets. Fuzzy sets are considered similar if
they are defined by overlapping membership
functions that assign approximately the same values
of membership to the elements in their universe of
discourse. Their similarity is the degree to which
they can be considered as equal.
In this work, in order to tackle the different
challenges of service selection, we propose a system
composed by mainly five steps as shown in Fig.1:
- First of all, users requirements on functional
criteria and non functional criteria are collected.
An initial set containing the deployed services
that offer the required functionality with the
required Quality of Service is given.
- We compute the Skyline services of our search
space using fuzzy dominance relationship.
- We refine the output by determining the set of
services that correspond the most to the user's
preferences using Fuzzy Similarity Measures.
- Finally, we will select from the top-K services
for each service class, the services that will
guarantee a coherent execution according to
transactional properties.
Balancing between Local and Global Optimization of Web Services Composition by a Fuzzy Transactional-aware Approach
Figure 1: System Architecture.
3.1 Computing Service Skyline
The first step in our proposed approach is computing
the set of skyline points from the search space. This
step will enormously reduce the candidate services
as it will select only the non-dominated services.
Skyline can be formally defined as follows:
Given a set of points S in a space with D
dimensions, Skyline points are the points who are
not dominated by any other point in the search space
according to those dimensions. A definition of the
dominance concept is then crucial to the
understanding of the skyline concept.
Pareto Dominance
the number of dimensions in the space
two web services in the space, we say that
denoted by
is at least
as good as
in all the dimensions and strictly
better in at least one dimension.
Illustrative Example
Let’s consider the common example in the literature
that selects the set of interesting hotels in a
reservation service represented in Fig.2. The hotels
are represented by two criteria: their prices and their
distances from the beach. It is obvious that a hotel
with a low price and a small distance from the beach
is preferred in this case. According to this, the
Skyline points are
as they are the only
points that are not dominated by any other point in
the search space.
Figure 2: Example of Skyline Set.
Fuzzy Dominance
Given two points in a space with
d dimensions, we
can define the dominance relationship as follows:
deg ( )
d : The space dimensions (i.e. the QoS attributes in
our context)
: two points in the search space
(), ( )
mi m j
: The values of the
attribute for
: A fuzzy membership function that is defined
as follows:
0 if x-y
(, ) 1 if x-y +
In this paper, we will compute Skyline services with
fuzzy dominance relationship by the Fuzzy Branch
and Bround Algorithm proposed in Rhimi, Ben
Yahia and Ben Ahmed approach (2015). The
proposed algorithm is a two-phase algorithm. The
first phase consists in transforming the data points of
the search space into a consistent fuzzy model.
Branch and Bound Skyline is an algorithm suggested
by Papadias (2003) based on R-Tree structure
known for its efficiency and effectiveness in large
spaces. It is widely used to reduce the search space.
The second step is determining the Skyline points
with a Branch and Bound algorithm according to the
fuzzy dominance relationship.
ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends
3.2 Computing Similarity between
Users Preferences and Services
using Fuzzy Similarity Measures
In our work, we chose to model user’s preferences
with fuzzy sets as they are highly appropriated for
expressing vagueness and smoothness in human
linguistic terms. In fuzzy set and possibility
framework, similarity of items is computed based on
the membership functions of the fuzzy sets
associated to the item features. Many Fuzzy
Similarity Measures were introduced in literature
Wu Zhang and Lu (2014), Zenebe and Norcio
(2010). In our study, for services
, the
following similarity measures between
denoted by
(I , I )
and are defined as:
min( ( ), ( ))
(I , I )
max( ( ), ( ))
ik i j
ik i j
()* ()
(I , I )
ik i j
ik i j
(() ())
(I , I )
{(), ()}
ik i j
iik ij
Max I I
(I , I ) 1 * ( ( ) ( ))
(2* ( ) 1) (2* ( ) 1)
kj ik i j
ik ij
are respectively the fuzzy
membership degrees of the i
QoS attribute of
. These fuzzy set based similarity
measures are: fuzzy set theoretic in (3), cosine in (4),
proximity in (5), and correlation-like in (6).
The outcome of this step is a set of services that
correspond the most to the user’s constraints.
3.3 Transactional-aware Selection
In this section we will no more consider individual
service classes. We will look at the whole service
composition and try to insure a global optimal
service selection.The description of the behavior of a
web service composition is how atomic web services
can be realized in terms of interactions with each
other. In a composition where several WS
components interact, unexpected behavior of a
component WS can not only lead to failure but also
can bring negative impact on all participants of the
composition. The execution of a composite web
service therefore requires transactional properties so
that the overall coherence is insured. Based on the
transactional properties and transactional rules of
web services that have been defined in Hadded et al.,
(2010), we adopt the following definitions:
A web service is said to be retriable and we note
it 'r' if it is sure to complete after a finite number of
activations. It is said to be compensatable and we
note it 'c' if it offers compensation policies to
semantically undo its effects. Finally it is said to be
pivot and we note it 'p' if once it successfully
completes, its effects remain and cannot be
semantically undone.
Now, we give the generalization of these
properties for a composite web service (CWS) as
defined by the authors.
A CWS is atomic if once all its component WSs
complete successfully, their effect remains forever
and cannot be semantically undone. Besides, if one
component WS does not complete successfully, then
all previously successful component WSs have to be
compensated. we will note 'a' an atomic CWS . A
CWS is compensatable and we note it 'c' if all its
component WSs are compensatable.. An atomic or a
compensatable CWS is retriable 'r' if all its
components are retriable.
Finally, we define a Transactional Composite
Web Service as a CWS whose transactional
behavioral property is in {a, ar, c, cr.}
Hence, in order to insure a transactional
Composite Web Service, the following rules should
be considered in the selection process:
- A 'p' or 'a' WS can only be sequentially
composed with a 'pr', 'ar', or 'cr' WS and can only
be executed in parallel with a 'cr' WS.
- A 'pr' or 'ar' WS can only be executed in
sequential or
in parallel with a 'pr', 'ar', or 'cr' WS.
- A 'c' WS can be sequentially composed with any
transactional WS but can only be executed in
parallel with a 'c' or 'cr' WS.
- A 'cr' WS can be executed in sequential or in
parallel with any transactional WS.
In our approach, rules for insuring a successful
termination of web services will be injected in the
selection algorithm after the initial phases of local
user-based optimization. Depending on the
transactional properties of the WS we will select the
candidates that insure a successful termination of the
composition according to the rules. For sake of
simplicity we will only consider a sequential
Balancing between Local and Global Optimization of Web Services Composition by a Fuzzy Transactional-aware Approach
workflow for our algorithm. We will let the parallel
workflow for future work.
The description of the selection approach is
given in algorithm 2.
Algorithm 2: TPAC (Transactional Preference- Aware
- Lists of top-k ranked concrete services
for each abstract service class computed by
Algorithm 2.
- composition workflow
// IAC: initial abstract class
1. Select the first element of the list
of IAC
2. found=0; i=0;
4. For all the AC in the workflow:
5. If ( current WS is 'p')
6. While (list_ next_ AC not empty and
7.if (element i of list is 'p' or
8. found=1;
9. Else if ( current WS is 'pr'
or 'ar')
10. While (list_ next_ AC and
11. if (element i of list
is 'pr' or 'ar' or 'cr')
12. found=1;
13. Else if ( current WS
is 'cr' or 'c')
14.Select the first
element of list_ next_ AC
// can be composed in sequence with any
transactional WS
4.1 Dataset and Experimental Setup
In our experiments, we adopt a real-world Web
service dataset: WSDream dataset which is designed
for QoS prediction approaches for Web service
recommendation and adopted in many papers such
as the works Zheng et al., (2014)
The dataset contains QoS records of service
invocations on 5825 Web services from 339 service
users, which are transformed into two user-service
matrices: a response-time user-item matrix and a
throughput user-item matrix. For studying the
prediction accuracy, we divide the dataset into two
parts, one part as training set and the other part as
testing set. In order to carry the predictions, we
randomly remove entries from the training user-item
matrices. Different methods are employed for
predicting the QoS values of the removed entries.
The original values of the removed entries are used
as the expected values to study the prediction
accuracy. Each experiment is evaluated by 10 times
10-fold cross validation. The 339 service users are
divided into two groups: 300 randomly selected
training users and the rest as test users. In 10-fold
cross validation, the training users are randomly
partitioned into 10 subsets. The process is repeated
10 times to predict the missing QoS values of test
users based on QoS values of the corresponding
training users. For studying the time execution, we
suggest a scenario of a sequential workflow with
eight service classes, but we vary the number of
services and the transactional properties of the web
services. We measure the average execution time
required to solve the composition problem of twenty
4.2 Evaluation Metrics
The following metrics have been used in this study:
Statistical Accuracy Metric
We use Mean Absolute Error (MAE) metric to
measure the prediction quality of our method with
the different fuzzy similarity measures. MAE is
defined as the deviation of recommendations from
their true user specified values and given by:
being the predicted rating of the service,
actual rating of the service and
the number of
predicted ratings.
Recall, Precision and F1 Metrics
Recall is defined as the fraction of preferred items
that are recommended. Precision is defined as the
fraction of recommended items preferred by the
user. The F1-measure, which combines precision
and recall, is the harmonic mean of precision and
In this experiment, a preferred rating threshold is
predefined. The preferred services are the services in
the test set whose actual ratings are greater than the
preferred rating threshold. The recommended
services are the services whose predicted ratings are
greater than the preferred rating threshold. The
recall, precision and F1 are defined as follows.
ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends
preferred recommended
preferred recommended
2* *
recall precision
recall precision
4.3 Experimental Evaluation
Effect of Varying Fuzzy Similarity Measures on
Prediction Accuracy:
We propose to study the effect of the similarity
measure on prediction accuracy. For this purpose,
we will compare the precision, recall, F1 and MAE
of our approach when using fuzzy set theoretic,
cosine, proximity and correlation-like measures.
Figures 3 and 4 show that for precision and
recall, correlation-like method has the highest
prediction accuracy. However Fig. 6 show that it has
a high MAE. Proximity and cosine have stable
results for all metrics, but cosine has the lowest
MAE which proves its efficiency.
Figure 3: Precision of the different similarity measures.
Figure 4: Recall of the different similarity measures.
Effect of Varying Number of Services Per Class on
Time Execution:
We notice that the computation cost increases when
the number of candidate services increases which
can be explained by the number of possible
combinations. However, we notice that time
execution is still reasonable as there are different
steps to reduce the number of possible candidates.
Figure 5: F1 of the different similarity measures.
Figure 6: MAE of the different similarity measures.
Thus, it is clear that the utilization of user's
preferences for pruning the services in the search
space eases time-consumption during the generation
of match results in the injection of transactional
Figure 7: Execution time varying number of services.
In this paper, we proposed a new method for
combining local and global web services
Balancing between Local and Global Optimization of Web Services Composition by a Fuzzy Transactional-aware Approach
optimization. Local optimization is given by
including users preferences in the selection process.
In fact, we suggest an approach that firstly computes
the set of Skyline services using fuzzy dominance
relationships. Then we refine the result according to
the user's expressed preferences. Our approach first
computes the similarity between the skyline services
and the user request with fuzzy similarity measures.
Finally, global optimization represented in our
study by successful termination of the composition
is injected. We make sure that in the selection
process, only candidates that guarantee transactional
properties of web services are chosen. The results
showed improvements in accuracy. We also proved
that the approach is not time-consuming as many
steps are used to prune that candidates.
In future work, we will study the injection of
transactional properties in a parallel workflow with
all possible combinations. We will also try to
include global constraints on the QoS in order to
insure global optimization from another perspective.
Abourezk,M., A. Idrissi. "Introduction of an outranking
method in the Cloud computing research and Selection
System based on the Skyline", IEEE Eighth
International Conference on Research Challenges in
Information Science, 2014, pp. 1-12.
Almulla, M., K. Almatori and H. Yahyaou, "A QoS-Based
Fuzzy Model for Ranking Real World Web Services.
Web Services" 2011 IEEE International Conference
on Web Services, 2011, .pp. 203 – 210.
Alrifai, M. and T. Risse, 2009. Combining global
optimization with local selection for efficient qos-
aware service composition. In International World
Wide Web Conference.881–890.
Alrifai, M., D. Skoutas, and T. Risse," Selecting skyline
services for qos-based web service composition" in
Proceedings of the International World Wide Web
Conference, 2010, pp. 11-20.
Ardagna, D., B.Pernici, 2007.Adaptive Service
Composition in Flexible Processes. IEEE Trans.
Software Eng. 369-384.
Benouaret, K., D. Benslimane, A. Hadjali, "On the Use of
Fuzzy Dominance for Computing Service Skyline
Based on QoS", In the 9th International Conference
on Web Services (IEEE ICWS 2011), 2011, pp.540-
Bhiri, S., O.Perrin and C.Gaudart, "Transactional Patterns
for Reliable Web Services Compositions" in
Proceedings of the 6th International Conference on
Web Engineering, 2006, pp. 137-144.
Canfora,G., 2005.An approach for QoS-aware service
composition based on genetic algorithms. Proceedings
of the 2005 Conference on Genetic and Evolutionary
Computation. 1069-1075.
Chen, L., "Ensuring reliability and qos optimizing for web
service composition", Computational Intelligence and
Security Conference, 2014, pp. 510-513, 2014.
Haddad,J. E., M. Manoeuvrier and M.Rukoz,
"TQoS:Transactional and QoS-aware selection
algorithm for automatic Web service composition",
IEEE Transaction on Service Computing, vol. 3, pp.
73-85, 2010.
Liu,A., L.Huang and Q.Li, "QoS-Aware Web Services
Composition Using Transactional Composition
Operator" in Proceedings of Web-Age Information
Management Conference, 2006, pp. 217-228.
Papadias,D., Y. Tao, G. Fu, "Progressive Skyline
computation in database systems", Journal ACM.
Rhimi, F., S. Ben Yahia and S. Ben Ahmed, " Novel
Approach for Computing Skyline Services with Fuzzy
Consistent Model for QoS- based Service
Composition", International Joint Conference on
Software Technologies, 2015, pp. 135-143.
Torres, R., H. Astudillo, R. Salas, "Self-Adaptive Fuzzy
QoS-Driven Web Service Discovery", IEEE
International Conference on Services Computing,
2011, pp. 64–71.
Wang, P., K.M. Chao, C.C. Lo, C.L. Huang and Y. Li, "A
Fuzzy Model for Selection of QoS-Aware Web
Services", IEEE International Conference on e-
Business Engineering
, 2006, pp. 585-593.
Wu, D., G.Zhang, and J.Lu, "A Fuzzy Preference Tree-
Based Recommender System for Personalized
Business to Business E-Services", IEEE Transactions
on Fuzzy Systems, Vol.23, pp. 29-43, 2014.
Xuan, V. and H. Tsuji, "QoS Based Ranking for Web
Services: Fuzzy Approaches", In the 4th International
Conference on Next Generation Web Services
Practices, 2008, pp. 77 – 82.
Yu, T. and L.Kwei-Jay,Service selection algorithms for
Services web withend-to-end QoS constraints, 2004.
Lin e-Commerce Technology Proceedings IEEE
International Conference.129 – 136.
Zadeh, L. A.. "Fuzzy sets". Information and Control, vol.
8, pp. 338-353, 1965.
Zenebe, A. and A.F.Norcio," Representation, similarity
measures and aggregation methods using fuzzy sets
for content-based recommender systems", Fuzzy Sets
and Systems, vol. 161, pp. 3044-3063, 2010.
Zeng, L., B. Benatallah, M. Dumas, J. Kalagnanam, and
Q. Sheng, 2003. Quality-driven Web Service
Composition. In WWW 411-421.
Zheng, Z., Y.Zhang, and M.R. Lyu, "Investigating QoS of
Real-World Web Services", IEEE Transactions on
Services Computing, vol.7, pp.32-39, 2014.
ICSOFT-PT 2016 - 11th International Conference on Software Paradigm Trends