Absfracf - Web services play an active role in the business
integration and other fields such as bioinformatics. Current Web
services technologies such as WSDL, UDDI, BPEL4WS and
BSML are not semantic-oriented. Several proposals have been
proposed to develop Semantic Web services to facilitate the
discovery of relevant Web services [1][8]. In our vision, with the
mature of Semantic Web services technologies, there will be a lot
of public or private Semantic Web services Registries based on
specific ontologies. These Registries may provide a lot of similar
Web services. So bow to provide the high quality of service (QoS)
Semantic Web services for specific domain using these Registries
will be a challenge task. Different domains have different
requirements of QoS, it is impractical to use classical
mathematical modeling methods to evaluate the QoS of Semantic
Web services. In this paper, we propose a framework called Soft
Semantic Web services Agent (SSWSA) for providing high QoS
Semantic Web services using soft computing methodology. And
we will use fuzzy neural network with CA learning algorithm as
our study case. Simulation result shows that the SSWSA could
handle fuzzy and uncertain QoS metrics effectively.
I. INTRODUCTION
Web services are playing more and more important role in
e-business application integration and other fields such as
bioinformatics. So it is crucial to provide and invoke the high
quality of service (QoS) Web services for the success of both
service providers and service consumers. Unfortunately,
current Web services technologies such as WSDL, UDDI,
ebXML, XLANG, WSFL, BPEL4WS and BSML are not
semantic-oriented that means they do not provide semantic
information-implicit semantics of service requestor and
service designer about Web services during the publishing,
discovery and invocation processes.
Several projects are undergoing to try to add semantics to
the Web services framework. DAML-S from DAML Services
Coalition [I] uses DAML+OIL based ontology for describing
Web services. The latest draft release [5] of DAML-S uses
WSDL in conjunction with DAML-S for Web service
description. METEOR-S from LSDIS Lab in the Computer
Science Department of the University of Georgia [SI follows
the approach to relate concepts in WSDL to DAML+OIL
ontologies in Web service description and then provide an
interface to UDDI that allows querying based on ontological
concepts.
In our vision, with the mature of Semantic Web services
technologies, there will be a lot of public or private Registries
for hosting and querying Semantic Web services based on
specific ontologies. Just as today’s situation, there are many
public and private UDDI Registries hosting a lot of similar
Web services with different QoS. For example, XEMBL, BQS
and OmniGene. We think the QoS of Semantic Web services
should cover both functional and non-functional properties.
Functional properties include the input, output,
conditionaloutput, precondition, accesscondition and effect of
service. These functional properties can be characterized as
the capability of the service [I]. Mon-functional properties
include the availability, accessibility, integrity, performance,
reliability, regulatoly, security, response time and cost.
Several matchmaking schemes have been proposed to
match the service requestor’s requirement with service
provider’s advertisement [7][8][14[15]. Here we must be
aware that on the one hand, the degree of capability matching
and non-functional properties are all fuzzy and on the other
hand, different, domains have different requirements on nonfunctional
properties. It is difficult to use classical
mathematical modeling methods to evaluate the whole QoS.
Although there are several QoS models [3] existing, they are
not suitable for the case in this paper.
In this paper, we propose a fiamework called Soft
Semantic Web services Agent (SSWSA) to provide the high
QoS Semantic Web services based on specific ontology such
as gnome. The SSWSA itself is Semantic Web sercice. The
SSWSA comprises of six components: specific ontology
based Semantic Web services Registries crawler
(SOBSWSRC), specific ontology based Semantic Web
services repositoly (SOBSWSR), specific ontology based
Semantic Web services inquiry server (SOBSWSIS), specific
ontology based Semantic Web services publish server
(SOBSWSPS), soil Semantic Web services agent
communication server (SSWSACS) and intelligent inference
engine (LIE). The core of SSWSA is intelligent inference
engine. It uses soil computing methodology to evaluate the
whole QoS (both functional and non-functional properties) of
Sematnic Web services. In this paper, we use several specific
ontology based Semantic Web services for bioinformatics and
fuzzy neural network with GA algorithm as our study to show
the flexibility and reliability of soft computing methodology
for handling fuzzy and uncertain linguistic information. For
example, capability of service is fuzzy. It is unreasonable to
use crisp value to describe it. So we can use several linguistic
variables such as low or a little bit low and a little bit high or
high to express the concept of capability of service.
The paper is organized as follow. In section 11, we
present the necessary background of Semantic Web services
and soft computing methodology. Section 111 provides the
details of architecture of the SSWSA. Section IV gives the
design of fuzzy neural network with GA algorithm. Section V
provides the simulation results. And fioally, in section VI, we
conclude this paper and give the future direction.
11. BACKGROUND
In this section, we will give a brief overview of the
background knowledge related to this paper.
A. Semantic Web Services
Currently, the Web is just a collection of documents
which are human readable but not machine understandable. In
order to remedy this disadvantage, the concept of Semantic
Web is proposed to add semantics to the Web to facilitate the
information fmdiog, extracting, representing, interpreting and
maintaining. There are many Semantic Web technologies
available today such as: RDF, RDFS, SHOE, DAML-ONT,
OIL, DAML+OIL, DAML-L and OWL [13]. On the other
hand, industry proposes the Web services to transform the
Web from “passive state” - repository of static documents to
“positive state” - repository of dynamic services.
Unfortunately, the Web services standards such as WSDL,
UDDI, XLANG, WSFL, ebXML and BPEL4WS are all not
semantic-oriented. They are awkward for service discovery,
invocation and composition [8]. So it is natural to combine the
Semantic Web with Web services, the so-called Semantic
Web services. Several projects have been initiated to design
the framework for Semantic Web services such as DAML-S
[I] and METEOR-S [8].
B. Soft Computing Methodology
According to Zadeh [I I] “Soft computing differs from
conventional (hard) computing in that, unlike hard
computing, it is tolerant of imprecision, uncertainty, partial
truth, and approximation.” The principal constituents of soft
computing are fuzzy logic, neural networks and genetic
algorithms. More and more technologies will join into the soft
computing framework in near future. Fuzzy logic is primarily
concerned with handling imprecision; neural computing
focuses on simulating human being’s learning process; genetic
algorithms simulate the natural selection and evolutionary
processes to do randomized global search. Each component of
soft computing is complementary to each other. Using
combination of several technologies such as fuzzy-neural
system will generally get better solution.
III. ARCHITECTURE OF SSWSA
The Soft Semantic Web services Agent (SSWSA) can
provide high QoS Semantic Web services based on specific
ontology. SSWSA uses cenkdized dienuserver architecture.
But itself can also be implemented as a Semantic Web service.
The SSWSA comprises of six components: specific ontology
based Semantic Web services Registries crawler
(SOBSWSRC), specific ontology based Semantic Web
services repository (SOBSWSR), specific ontology based
Semantic Web services inquiry server (SOBSWSIS), specific
ontology based semantic Web services publish server
(SOBSWSPS), soft Semantic Web services agent
communication server (SSWSACS) and intelligent inference
engine (IIE). The high level architecture of SSWSA is shown
in Figure 1.
A. SOBSWSRC
The specific ontology based Semantic Web services
Registries crawler (SOBSWSRC) will access many public or
private specific ontology based Semantic Web services
Registries such as semantics-oriented UDDI for gnome to
fetch the service profiles and store them in specific ontology
based Semantic Web services Repository (SOBSWSR).
B. SOBSWSR
The specific ontology based Semantic Web services
Repository (SOBSWSR) will store the service profiles which
are fetched by SOBSWSRC and Semantic Web services
which are published by services provider through the
SOBSWSPS.
C. SOBSWSRC
The specific ontology based Semantic Web services
Inquiry Server (SOBSWSIS) will provide two kinds of
interfaces: one is the programmatic API to other agents;
another is the human-readable. The SOBSWSIS transforms
the service request profile into standard format and does
matchmaking operation to calculate the capability of Semantic
Web services stored in the SOBSWSR.
D. SOBSWSPS
The specific ontology based Semantic Web services
Publish Server (SOBSWSPS) will provide the API for service
providers to publish their specific ontology based Semantic
Web services into the SOBSWSR.
E. SSWSACS
The soft Semantic Web services Agent Communication
Server (SSWSACS) will use certain communication protocol
such as KQML and SOAP to communicate with other
SSWSAs. For example, if the SSWSA could not fmd the
appropriate Semantic Web services requested by the service
requesters, it will transfer the request to other SSWSAs
through the SSWSACS.
F. IIE
The Intelligent Inference Engine (IIE) will use soft
computing methodology to calculate the QoS of the Semantic
Web services.
IV. DESIGN OF FUZZY NEURAL NETWORK WITH GA
ALGORITHM
A schematic diagram of the four-layered fuzzy neural
network (FNN) is shown in Figure 2. Nodes in layer one are
input nodes representing input linguistic variables. Nodes in
layer two are membership nodes. Each membership node is
responsible for mapping an input linguistic variable into a
possibility distribution for that variable. The rule nodes reside
in layer three. The last layer contains the output variable
nodes [4].
Yt Y,
4 4
Layer 4
(output layer)
Layer 3
(rule layer)
Layer 1
(membership
layer)
As we mentioned before, metrics of QoS of Semantic Web
services are multidimensional. For illustration of specific
ontology based Semantic Web services for bioinformatics, we
decide to use capability, response time and trustworthiness as
our inputs and QoS as output. The fuzzy logic system (FLS) is
based on TSK model.
A. Design of Fuzzy Input Sets
Let x represent capability, y represent response time and z
represent trustworthiness. We scale capability, response time
and trustworthiness to [0, IO] respectively. The graphical
representation s of membership functions of x, y, z are shown
in Figure 3.
4
0 2 4 t 8 IO x, Y. z
Fig. 3. Membership Functions of Inputs
B. Design of Fuzzy Rule Base
Here, we design the fuzzy rule base based on TSK model.
Fuzzy rule is the form shown below:
If x is 1, and y is I2 and z is I3 THEN
0 is a,, * x + &> * y + aiz * z + &.+
where 11, I2 and 1, in {low, middle, high], I in [I, 271.
There are totally 27 fuzzy rules. The aij's are consequent
parameters which will he obtained by training phase of FNN
using GA algorithm.
C. Design OfDefrrrrifier
Suppose, for certain input x, y and z, there are m fuzzy
rules fxed. To calculate firing strength of jth rule, we use
below formula:
(1)
So the output is:
wi = P h ) * PAY) * P M
0 = &I ,... m W'*(q,l*x + q,2*Y + q,3*z + q.4) /(&=I ..... m wi)(2)
D. Design of Genetic Algorithm
GA is a model of machme learning which derives its
behavior from a metaphor of the processes of evolution in
nature. This is done by the creation within a machine of a
population of individuals represented by chromosomes. Here
we use real-coded scheme. Given the range of parameters
(coefficients of linear equations in TSK model), the system
uses the derivate-free random search-CA to leam to find the
near optimal solution by the fitness function through the
1)Chromosome: the genes of each chromosome are 81 real
numbers (there are 81 parameters in fuzzy rule base) which
are initially generated randomly in the given range. So
each chromosome is a vectnr of 81 real numbers.Z)Fitnesss&ncfion: The fimess function is defme as
3)Elitism: The tournament selection is used in the elitism
process.
4)Crossower: The system will randomly select the two
parents among the population and then randomly select
the number of cross points and simply exchanges the
corresponding genes among these two parents to generate
new generation.
5)Mufafion: For each individual in the population, the system
will randomly select the genes in the chromosome and
replace them with randomly generated real number in the
given range.
E = l/ZEi=i, ...,,,, (dj - 03' (3)
VI. SIMULATION
There are two phases for applying fuzzy neural network
(F"): training and predicting. In training phase, we use 168
data entries as training data set. Each entry consists of three
inputs and one expected output. We tune the performance of
the system by adjusting number of population, generation and
probability of crossover and mutation. Table I gives the part
of prediction result with several parameters for output 0.
TABLE I PREDICTION RESULT OF FNN.
I 5.4
In Table I, NO. OF GENERATION = 1000000, NO. OF
POPULATION = 1000, PROBABILITY OF CROSSOVER =
0.7, PROBABILITY OF MUTATION = 0.1. From the Table
I, we could see the maximum error of training result is 1.6.
The total prediction error for 168 entries is 25%. By our
observation, designing reasonable fuzzy membership
functions and choosing reasonable training data set which is
based on specific application domain can reduce the
prediction error a lot.
VI. CONCLUSIONS
In this paper, we discuss the design of Soil Semantic Web
services Agent (SSWSA). The SSWSA could be used in
World Wide Web (WWW), P2P and Grid intYastructure. The
SSWSA is flexible and extensible. With the evolution of softcomputing, more and more technologies can be integrated into
SSWSA. We use specific ontology based Semantic Web
services for bioinformatics and fuzzy neural network with
genetic algorithm as OUT study case. The trainiig time is short
and training result is satisfactory. The SSWSA will retum the
desired Semantic Web services based on the QoS of Semantic
Web services. In the future, we will implement the SSWSA
with the mature of Semantic Web services technologies, and
apply it to semantic Web-based bioinformatics applications.
ACKNOWLEDGEMENT
The authors would like to appreciate the support by NIH
under Grant P20 GM065762. We thank Dr. Harrison for his
support and MI. Tang for his valuable suggestions.
REFERENCES
[I] A. Ankolekar, M. Burstein, J. Hobbs. etc. DAML-S: Web Senice
Description for the Semantic Web The Fiml Intemalioml Semantic Web
Conference (SWC), Sardinia (Italy), June. 2002.
[2] A. Mani, and A. Magarajan Understanding quality of senice for Web
senices. h t t p : / ~ - l 0 6 . i b m . c o m / d e v e l o p e M . o r k s / l i b h h l
[2] A. Sheth, J. Cardoso, I. Miller and K. Kwhut QoS for Service-oriented
Middleware Proceedings of the Conference on Ystemics, Cybemetics and
In/omotics, Orlando, FL, July 2002 .
[3] Ching-hung Lee, Jam-Lee Hong, Yuching Lin and Wei-yu Lai Type-2
Fuzzy Neural Newark Systems and Leaning. Inremotional Journal of
Computational Cognition Vol. 1, No. 4, pp. 7 S 9 0 , December 2003.
[SI DAML-S 0.7 Dratl Release h r m : / ~ . d o m l . o r e / s o i ~ ~ / d e m l - s / 0 . 7 /
[6] Herve Abdi A Neural Newark Primer. Joumal U/ Biological System,
Vol.2, No. 3, pp. 247-283, 1994
[7] J. G o d c z -Castillo, D. Trastour, C. Bartolini Description Logics for
Matchmaking of Services w?uwh~l.ho.com/rechreoorrs/2OOl/HPL-2OOl-
&l.�L&
[SI K. Sivashanmugam, K. Verma, A. Sheth and I. Miller Adding Semantics
to Web Services Standards Intemafionol Conference on Web Ssnices (7CWS'
03), 2003.
[9] K. F. Man, K.S. Tang and S. Kwoog Genetic Algorithms: Concepts and
Applications. IEEE Trmx On Industrial Electmnics, Vol. 43, No. 5, October
1996.
[IO] L.A. Zadeh The concept of a linguistic variable and its application to
approximate reasoning-I, Inform. Sci. Vo1.8, pp. 19%249, 1975.
[ I l l L.A. Zadeh Fuvy logic, neural networks, and soft computing.
Communicalionr of the ACM, Vol. 37, pp. 11-84. 1994.
[I21 M. Srinivas and Lalit M. Pamaik Genetic Algorithms: A Survey.
Computer, Vol. 21, pp. 1 7 4 6 , 1994.
[I31 S.A. Mcllmith, T.C. Son and H.L. Zeng Semantic Web service^ IEEE
lnlellizent Svstems. S~ecinl Issue on the Semantic Web. 16(2):4653. ,,
March/April,'2001.
[I41 Sycara, K., Khuch, M., Widoff, S. and Lu, I. Dynamic Senicc
Matchmaking Among Agents in Open Information Enviro&enfs. SIGMOD
Record (ACMSIG on ManagemenlofDalo) Val. 28 No.1, pp47-53, 1999.
[IS] T. R. Payne, M. Paolucci, KSycara Advertising and Matching DAML-S
%Nice Descriptions hnp
:/hvwwsemantioueb. o ~ g / s W i t i ~ ~ ~ ~ i ~ ~ "
[返回] 中文XML论坛 - 专业的XML技术讨论区 →
W3CHINA.ORG讨论区 - Web新技术讨论 → 『 Semantic Web(语义Web)/描述逻辑/本体 』 → Soft Semantic Web Services Agent【一篇论文】
(订阅本版)
2006/5/9 13:42:00
新浪微博
Soft Semantic Web Services Agent【一篇论文】
