A GRAPH-BASED APPROACH FOR WEB SERVICES

COMPOSITION

Salima Benbernou, Djamal Benslimane, Mohand Said Hacid

LIRIS, Computer Science department, Claude Bernard Lyon 1 University, France

Hamamache Kheddouci

LIESP, Claude Bernard Lyon 1 University, 43 boulevard du 11 Novembre 1918, 69622 Villeurbanne, France

Kamel Tari

omputer Science Department, Bejaia University, Algeria

Keywords:

Oriented graph, semantic web services, web services composition, web services discovery.

Abstract:

The automatic web services composition plays an important role in the area of semantic web services. In fact,

in a real life it is not usual to discover atomic web services matching with the user’s query. So composing

services fulﬁlling the query is needed. This paper discusses the web services composition model based on an

oriented graph. It is based on Bcov algorithm (Benatallah.B et al., 2005a), initially deﬁned for web services

discovery and considered as the best cover problem, which it was mapped as the problem of computing the

minimal transversals with minimum cost of a weighted hypergraph. We propose two algorithms handling

the oriented graph reasoning for web service composition. At ﬁrst glance, a basic composition version for

Bcov algorithm is given. The algorithm composes services over the induced subgraphs of each transversal

provided by Becov algorithm, and gives a covering ratio satisfying the query. The second algorithm used a

composed services graph maintained by a web services provider. The composition in that case is based on

Dijkstra algorithm, where for each two consecutive services in the best minimal transversal provided by Bcov

algorithm, a shortest path is computed regarding the composed services graph.

1 INTRODUCTION

Many organizations are putting their business avail-

able through the Internet as Web services. One of

main challenge issue in the area of semantic web ser-

vices is the automation of the combination of these

services. Therefore, there is a need of tools devel-

oping complex services by composing other services

with total transparency.

There exists a plethora of different approaches to

Web service composition and for that reason a corre-

sponding number of summaries and overviews as well

(Hull et al., 2003; Hull, 2003; Rao and Su, 2004).We

limit ourselves of some of them.

Composition realized by BPEL4WS rests on a static

workﬂow level without taking any semantic informa-

tion into account (Yuan.Y et al., 2006). Hereafter, we

present recent practical and theoretical approaches to

automated composition.

In (Benbernou.S et al., 2004b; Benbernou.S et al.,

2004a) an hybrid framework combining DL ontolo-

gies with Constraint Satisfaction Problem (CSP) is

presented. This approach utilizes a calculus reason-

ing about a DL terminology and subsequently solv-

ing constraints imposed on the given services. The

proposition in (Bultan et al., 2003) is expressed by the

means of the conversation notion. They introduce a

framework for modeling and specifying the global be-

havior of Web service composition, in which the peers

(individual Web service) communicate through asyn-

chronous messages. A global watcher keeps track of

messages as they occur. A conversation is in this con-

text a sequence of (classes of) messages observed by

the watcher. The peers are represented also by Mealy

Finite State Machines (MFSM)(John E. Hopcroft, R.

Motwani and Jeffrey D. Ullman, 2001).

(Hamadi and Benatallah, 2003) propose a full algebra

of Petri-Nets

. Each Web service is represented by

A Finite State Machine (FSM) (John E. Hopcroft, R.

290

Benbernou S., Benslimane D., Said Hacid M., Kheddouci H. and Tari K. (2007).

A GRAPH-BASED APPROACH FOR WEB SERVICES COMPOSITION.

In Proceedings of the Third International Conference on Web Information Systems and Technologies - Internet Technology, pages 290-296

DOI: 10.5220/0001289102900296

 SciTePress

such a net including the null net. Operations are for

instance sequence, alternative or execution in parallel

of two services. However, all existing constraints are

implicit within the transitions of the Petri-Nets. In this

approach, they only take the behavior into account de-

scribed by a sequence of actions and messages. In

(Berardi.D et al., 2003a; Berardi.D, 2004; Berardi.D

et al., 2003b) the authors relate an e-service execution

to an internal and an external schema. These schemes

correspond respectively to an unfolded internal and

external FSM describing the behavior of the service

in terms of actions. They do not consider any con-

straints on inputs/outputs. (Bordeaux.L et al., 2004)

constitutes another approach concentrating on com-

position between two services in a behavioral context.

Web service composition can be considered as a form

of Artiﬁcial Intelligence (AI) (Stuart J. Russell and P.

Norvig, 2003) planning task.

In (McIlraith and Son, 2002; Narayanan and McIl-

raith, 2002) composition is addressed by using the

language GOLOG and providing a behavioral seman-

tic of the composition based on Petri Nets (for OWL-

S). GOLOG is a logic programming language based

on situation calculus(R. Reiter, 2002). The authors

treat the Web service composition problem through

the provision of high level generic procedures and

customizing constraints. GOLOG is adapted for rep-

resenting and reasoning about this problem. In (Med-

jahed et al., 2003) the authors present a technique to

generate composite services from high level declara-

tive description. The method uses composability rules

to determine whether two services can be copmposed.

There are syntactical and semantical rules: the syn-

tactical ones for protocol bindings and the seman-

tical rules to manage 1) message composability, 2)

operational semantic composability and composition

soundness.

In (Doshi.P and Zhao.H, 2006), the authors present

a hierarchical approach for composing web processes

that may be nested some of the components of the pro-

cess may be Web processes themselves. They model

the composition problem using semi-Markov decision

process (SMDP) that generalizes MDPs by allowing

actions to be temporally extended. They use these ac-

tions to represent the invocation of lower level Web

processes whose execution times are different from

simple service invocations.

In (Tan.M et al., 2006) is proposed an agent-based

method using Fuzzy Distributed Constraint Satisfac-

tion Problem (Fuzzy DisCSP). They provide tech-

niques to solve the problem of QoS composition prob-

Motwani and Jeffrey D. Ullman, 2001) allows to verify eas-

ily properties of a composition - often using model checking

(Merz, 2000) methods for example.

lem, which is how to compose a service from differ-

ent subcomponent services so that its overall QoS can

satisfy certain requirements. The authors show that

by using the composition structures, local constraints

can be constructed and used with DisCSP.

In this paper we propose an oriented graph based

model for modeling web services composition upon

Bcov algorithm. The framework allows to compose

services by providing two approaches:

1. by means the services discovered by Bcov al-

gorithm, ﬁnds the compositions as paths (or

subgraphs) between two services regarding the

matching of the outputs with inputs,

2. Dijkstra’s algorithm is based on our calculus for

the composition to ﬁnd the shortest Path between

two services given a global oriented graph main-

tained by a web services provider.

In the rest of the paper, we shall not handle the seman-

tic matching between inputs and outputs, we assume

it is computed in a black box.

The remainder of the paper is organized as follows:

The next section gives two motivating examples for

the proposed approaches. In the section 4 we intro-

duce the oriented graph model for web services com-

position and provide a reasoning approach by looking

for a path or a subgraph fulﬁlling the user query. In

Section 5, due to the limitations of the previous algo-

rithm, we provide a second approach using a global

oriented graph which is managed by a web services

provider (for web services composition) and ﬁnding

the shortest path for the composition of services.

2 MOTIVATING EXAMPLES

In order to show the interactions between web ser-

vices in terms of inputs and outputs, let us illustrate

our approach through examples.

Example 1 Given a user query looking for a service

translator providing the translation text from French

to Thai language, the query input is French and the

output is Thai. let us assume a web services provider

describing a set of following services:

1. Translation from French to English (S

F −E

)

2. Translation from English to German (S

E−G

)

3. Translation from English to Chinese(S

E−C

)

4. Translation from Chinese to Thai ... (S

C−T

)

A web service fulﬁlling the query constraints is not

discovered. However, if we consider all services, we

can ﬁnd a composition of them, answering the query,

given by the following ordered sequence {(S

F −E

A GRAPH-BASED APPROACH FOR WEB SERVICES COMPOSITION

291

E−C

), (S

C−T

)}. This composition is called lin-

ear. This composition is made through the matching

between outputs and inputs of services.

Example 2 Let us consider the client query search-

ing for a travel accommodation. Let us assume a web

services provider, giving following web services:

• Tour-club, having a client name and credit

card as input providing Flight-reservation and

Accommodation-reservation.

• Leisure-club, having Accommodation-

Reservation as inputs providing swimming-leisure

• Renting-car, having Flight-reservation as input

providing (Renting-car,Renting-boat).

The provider could not ﬁnd an atomic services ful-

ﬁlling the constraint of the requester, so the answer is

a composition of following web services: Tour-club

+(Leisure club + renting car), this composition is

called non linear.

3 WEB SERVICES

COMPOSITION

Informally speaking, each services is deﬁned by its

inputs and outputs.

3.1 Basic Deﬁnitions

Deﬁnition 1 (A Web service.) A Web service S is de-

ﬁned as a pair (I, O) where I = {I

| i > 0} a set of

service inputs and O = {O

| j > 0} a set of service

outputs.

Deﬁnition 2 (Semantic composition.) Two services

et S

can be composed semantically together

which we denote by S

⊲ S

iff ∃O

∈ O(S

), ∃I

∈

I(S

) | O

⊑ I

where O(S

) and I(S

) are sets

of service outputs and inputs respectively, and ⊑ de-

notes a semantic matching symbol.

The deﬁnition means two services can be com-

posed iff the outputs (not necessarily all) of one match

semantically the inputs of the other. We assume the

matching is doing through a black box and is done

using existing tools as those presented in (Sycara.K

et al., 2003; Benatallah.B et al., 2005a).

Deﬁnition 3 (Atomic web services.) A service S

called atomic iff 6 ∃(S

⊲ S

), where S

6= S

Deﬁnition 4 (A web services oriented-graph.) Let G

be an oriented Graph deﬁned by (S, E)where, S is a

set of web services deﬁned by its inputs and outputs,

and E a set of oriented edges, such as an ongoing

edge denoted by

−→

, S

) means a service S

can be

composed with S

regarding the inputs and outputs.

Deﬁnition 5 (A linear composition.) Let G be an

oriented graph deﬁned by (S, E), the composition is

called linear iff for each node there is only one outgo-

ing edge, ∀S

, S

∈ S, ∃e ∈ E | e =

−→

, S

), j ≥ 1

and |e| = 1, wher e |(S

, Sj)| means the cardinality

of the edges between S

, S

. For instance in Figure

1(a), all services have only one outgoing edge.

Deﬁnition 6 (A non linear composition.) Let G be

an oriented graph deﬁned by (S, E), the composition

is called non linear iff for each node there is more

than one outgoing edge, ∀S

, S

∈ S, ∃e ∈ E |

e =

−→

, S

), |e| > 1. For instance in Figure 1 (b),

the service S

has two outgoing edges.

(b) Non linear composition

(a) Linear composition

Figure 1: Examples of web services composition graph.

3.2 Overview of Bcov Algorithm

In (Benatallah.B et al., 2005b; Benatallah.B et al.,

2005a), an algorithm for web services discovery was

provided. It is a hypergraph based approach. The

key idea underlying the Becov algorithm is: given

a query Q and a set of services S, the problem of

discovery consists in ﬁnding a subset of S called a

”cover” of Q that contains as much as possible of

common information with Q and as less possible of

extra information with respect to Q. The services and

query are described by a conjunction of atoms (in de-

scription logic formalism). The services are discov-

ered through the construction of a hypergraph (Σ, Γ)

where Σ is a set of vertices corresponding to services

description and each conjuncts in the query becomes

an edge of Γ in the hypergraph. The algorithm com-

putes the minimal transversals with the best cost for

the hypergraph looking for a best cover of services

minimising the conjunction of services matching the

maximum of atoms in the query (for more details

see (Benatallah.B et al., 2005b; Benatallah.B et al.,

2005a).

Let us illustrate the algorithm by the following exam-

ple, let us assume we have a query deﬁned by the fol-

lowing conjunction Q = A

∧A

∧

WEBIST 2007 - International Conference on Web Information Systems and Technologies

292

∧ A8 ∧ A9 and a set of services S

, S

, ....S

In Figure 2 is depicted the corresponding hypergraph

and the best cover of the query Q, represented by two

minimal transversal: (S

∧ S

), (S

∧ S

S1 S2

2 Minimal

transversals

Empty

edge

A1 ∧ A4

A3 ∧ A5

A7 ∧ A6

Figure 2: Example of web services hypergraph modeling.

3.3 Oriented Query

Deﬁnition 7 (oriented query.) A query Q is called

oriented and denoted by

−→

iff the atoms composing

Q are ordered.

Example 3 Let us consider the following query . ”I

want to go from Paris to Barcelona on 3rd March,

look for an accommodation there for one week (from

3rd to 10th of March), and rent a car”. We can ﬁnd

a ﬂight or railway services and hotel service and so

on. However, we can not accept what the service ho-

tel offers me before ﬁnding ﬂight service . So the dis-

covered services might respect an order to fulﬁll the

query.

Such reasoning is not provided in Bcov algorithms.

So next, we will show how we can improve it by in-

troducing the composition .

3.4 Basic Reasoning with Bcov for the

Composition

The idea behind the composition which we propose is

given by the following steps:

• for each minimal transversal provided by Bcov al-

gorithm, we build an oriented graph where each

node corresponds to a service and the edge are the

relationships between inputs and outputs of ser-

vices (in the transversal).

• after discovering atomic services which are iso-

lated nodes in the graph, we order the services

of the minimal transversal regarding the query

atoms.

• atomic services are removed from the transversal.

And we try to ﬁnd in the oriented graph, a path

(a linear composition) or a subgraph (non linear

composition) between ﬁrst and the last services

(of the ordered set) taking into consideration all

the intermediary services.

• we compute the cost of the retrieval by consid-

ering only the composed services regarding the

query. The cost is given by the covering ratio.

Deﬁnition 8 (Covering ratio) It is computed by the

ratio of the number of web services induced in the

composition on the total of services provided in the

best cover set: CR =

|composed−web−services|

|total−services|

, where

|| means the cardinality.

Let us illustrate the approach by the following exam-

ple:

Example 4 Let us assume, a query described by

the following atoms:

−→

and a set of minimal transver-

sal provided by Bcov algorithm:{T

, T

}

where T

= {S

, S

}, T

, S

= {S

, S

}. For T

, T

, we

build a corresponding graph where the inputs match

the outputs. Figure3 gives an example of the oriented

graph associated to T

. We notice in the graph it may

exist a nonlinear composition for S

Figure4(a)shows the mapping between query atoms

and the services provided in T

. Figure4(b) provides

the ordered set after the mapping.

Finally, the aim is to ﬁnd a path (or a subgraph

) starting from S

to S

including S

, S

the oriented graph (S

is an atomic service). The

only path we can ﬁnd is a subgraph which is a

nonlinear composition (S

, S

) and (S

, S

) de-

picted in Figure5. We now compute the covering

ratio:CR =

(3+1)

. The same calculus is done

for all minimal transversal and at the end we chose

the one with the best ratio.

S1 S3

Figure 3: oriented graph for T

transversal.

A GRAPH-BASED APPROACH FOR WEB SERVICES COMPOSITION

293

Q: A1,A2, A3, A4, A5, A6, A7

T1: S1, S3, S6, S7, S4, S8

T1: S1, S6, S3, S7, S4, S8

(a)

(b)

Figure 4: (a)atoms belonging to the best cover service T1,

(b) best cover ordered regarding the query atoms.

S1 S3

Well composed services

regarding the query

Figure 5: Well composed services in the oriented graph.

The Bcov improvement is given by Bcovcomp algo-

rithm.

3.5 Discussion About Bcovcomp

Statistically speaking, the algorithm can be executed

for at most 20 services for each minimal transver-

sal, consequently the composition can be processed

by hand. Furthermore, the covering ratio is used for

many possibilities of the composition. The complex-

ity is reasonable (of course after applying Bcov which

is NP-hard) it is linear. However, as we see in the ex-

ample, we can not ﬁnd a path from S

to S

. This

problem arises due to the fact that the graph is local,

where the composition is limited only for services in

the transversal. To overcome the problem, in the next

section we give an approach based on a global ori-

ented graph maintained by a provider.

4 ORIENTED-GRAPH

REPRESENTATION FOR WEB

SERVICES COMPOSITION

We assume a web services provider maintains an

oriented-graph called global graph where each ver-

tex represents a service and an outgoing edge means a

possible composition between two services published

by the provider (an output of the service with the in-

put of some services). The idea behind the second

approach is, considering the transversal minimal with

the best cost (not using all minimal transversals as

Algorithm 1 Bcovcomp. (sketch)

Require: A set BS: the minimal transversales of the

query Q provided by BCOV.

Ensure: A set WBS: the well composed of the best

covers of the query Q.

1: for all set of the covers SC

∈ BS do

2: Built an oriented graph OG

for SC

3: order SC

regarding the order of query atoms.

4: for all S

∈ SC

do { // Looking for atomic web

services}

5: if S

is atomic then

6: the service can be considered in any order

in the best cover.

7: else

8: It can be composed, keep it.

9: end if

10: end for

11: Find a path (for linear composition, or a sub-

graph for non linear composition) stating from

the ﬁrst service to the end one in OG.

12: Compute the covering ratio CR

13: end for

14: Choose the best cover regarding the best cover-

ing ratio.

we did in the previous algorithm), ﬁnding the short-

est path between two consecutive services of the or-

dered transversal (regarding the oriented query, with-

out considering the atomic services). At the end of

the process, we obtain a subgraph which is consid-

ered the solution of the query. Let us ﬁrst recall the

shortest path problem.

Deﬁnition 9 (The shortest path problem.). It is de-

ﬁned as follows: Given a graph G = (S, E), and

e ∈ E, we associate to each edge a weight l(e) ≥ 0

(in our case we assume all edge have a weight equal

to 1) we call it the length of the edge u. Find an ele-

mentary path µ , starting from an edge S

to another

one S

, such as the total length l(µ) =

e∈µ

l(e) is

as small as possible.

There are many algorithms handling the shortest

path problem, we use the one developed by Dijk-

stra algorithm (for more details of the algorithm see

(E.J.Dijkstra, 1959; S.Skiena, 1990), we denote it by

Dijkstra(G,s), which given a graph G ﬁnds all shortest

paths from s to each other vertex in the graph. And

shortestPath(G,s,t) uses Dijkstra to ﬁnd the shortest

path from s to t.

Example 5 Let us illustrate the second approach

by considering the previous example: let us assume

that the transversal T

is the best cover of minimal

transversal and the provider maintains a dynamic

global oriented graph for the composition of web ser-

vices depicted in ﬁgure6. The aim now is to ﬁnd a path

or a subgraph for each consecutive services of

−→

WEBIST 2007 - International Conference on Web Information Systems and Technologies

294

shortestPath(G, S

, S

), shortestPath(G, S

, S

shortestPath(G, S

, S

), shortestPath(G, S

, S

We consider S

as atomic service. At the end of the

calculus we obtain a global subgraph fulﬁlling the

query by composing services not provided by Bcov.

So the subgraph induced between S

and S

is a path

deﬁned by

−→

, S

) (it is a linear composition).

We will do the same process for the remainder of the

pair of services. The obtained subgraph is depicted

in Figure 7. Two services S

and S

was introduced

in the composition beside the services provided by

Bcov algorithm. A sketch of the algorithm is given by

Algorithm 2.

Figure 6: The global oriented graph.

Figure 7: The composition fulﬁlling the query.

Algorithm 2 Shortest Composition Algorithm

SCA. (sketch)

Require: The minimal transversal provided by Bcov:

The global oriented graph GOG

Ensure: A subgraph meaning a possible composition

including all services involved in the transversal

minimal.

1: for all two consecutive services S

, S

i+1

∈ T do

2: Apply shortestP ath(GOG, S

, S

i+1

))

3: end for

4: Provide the global subgraph for all services in-

duced in the composition of services fulﬁlling the

query.

Complexity

The worst-case running time for the Dijkstra algo-

rithm on a graph with n nodes and m edges is O(n

Where O(n

) for node selection and O(m) for dis-

tance updates. While is the best possible complex-

ity for dense graphs, but can be signiﬁcant improved

upon for sparse graphs.

5 CONCLUSION

In this paper we have investigate the problem of

the web services composition by providing two ap-

proaches based on graph theory, built upon Bcov al-

gorithm initially provided for dynamic web services

discovery. The ﬁrst one uses all the minimal transver-

sals provided by Bcov algorithm and tries to ﬁnd the

best one meaning a covering ratio and local oriented

graph. However, the second one uses only the min-

imal transversal with the best cost and tries to ﬁnd

the shortestPath between all ordered services (regard-

ing the query)in the global graph maintained by the

web services provider. The result obtained are very

interesting but still preliminaries. Taking only in-

put/output into account as features to build a compo-

sition relation is not strong and sufﬁcient. That’s why

our ongoing work is devoted to the extension of the

proposed framework to hold the constraints on the in-

puts/outputs in the graph by means a cost in order to

relax the composition.

REFERENCES

Benatallah.B, Hacid.M.S, Leger.A, Rey.C, and Toumani.F

(2005a). On automating web services discovery.

VLDB Journal, 14(1):84–96.

Benatallah.B, Hacid.M.S, Paik.H, Rey.C, and Toumani.F

(2005b). Towards semantic -driven, ﬂexible and scal-

able framework for peering and querying e-catalog

communities. Information Systems International

Journal.

Benbernou.S, Canaud, and Pimont.S (2004a). A seman-

tic web services discovery regarded as constraint sat-

isfaction problem. In 6th Flexible Query Answering

Systems FQAS’04.

Benbernou.S, Canaud.E, Hacid.M.S, and Toumani.F

(2004b). Resolution and constraint propagation for

semantic web services discovery. In RIDE, pages

23–30.

Berardi.D (2004). Automatic composition of ﬁnite state e-

services. Technical report, American Association for

Artiﬁcial Intelligence (AAAI).

Berardi.D, Calvanese.D, Giacomo.G, D., Lenzerini.M, and

Mecella.M (2003a). Automatic composition of e-

A GRAPH-BASED APPROACH FOR WEB SERVICES COMPOSITION

295

services that export their behavior. In ICSOC, pages

43–58.

Berardi.D, Calvanese.D, Giacomo.G, D., Lenzerini.M, and

Mecella.M (2003b). A foundational vision of e-

services. In WES, pages 28–40.

Bordeaux.L, Sala

un.G, Berardi.D, and Mecella.M (2004).

When are two web services compatible? In TES,

pages 15–28.

Bultan, T., Fu, X., Hull, R., and Su, J. (2003). Conversation

speciﬁcation: a new approach to design and analysis

of e-service composition. In WWW, pages 403–410.

Doshi.P and Zhao.H (2006). A hierarchical framework

for composing nested web processes. In 4th Inter-

national Conference on Service Oriented Computing

ICSOC’06.

E.J.Dijkstra (1959). A note on two problems in connection

with graphs. Numerische Math, 1:269–171.

Hamadi, R. and Benatallah, B. (2003). A petri net-based

model for web service composition. In ADC, pages

191–200.

Hull, R. (2003). E-service composition: Models and for-

malisms. In Description Logics.

Hull, R., Benedikt, M., Christophides, V., and Su, J. (2003).

E-services: a look behind the curtain. In PODS, pages

1–14.

John E. Hopcroft, R. Motwani and Jeffrey D. Ullman

(2001.). Introduction to automata theory, languages,

and computation. Addison-Wesley.

McIlraith, S. A. and Son, T. C. (2002). Adapting golog for

composition of semantic web services. In KR, pages

482–496.

Medjahed, B., Bouguettaya, A., and Elmagarmid, A. K.

(2003). Composing web services on the semantic web.

VLDB J., 12(4):333–351.

Merz, S. (2000). Model checking: A tutorial overview. In

MOVEP, pages 3–38.

Narayanan, S. and McIlraith, S. A. (2002). Simulation, ver-

iﬁcation and automated composition of web services.

In WWW, pages 77–88.

R. Reiter (2002.). Knowledge in Action. The MIT Press.

Rao, J. and Su, X. (2004). A survey of automated web ser-

vice composition methods. InSWSWPC, pages 43–54.

S.Skiena (1990). ”dijkstra’s algorithm.” 6.1.1. Implement-

ing Discrete Mathematics: Combinatorics and Graph

Theory with Mathematica, pages 225–227.

Stuart J. Russell and P. Norvig (2003.). Artiﬁcial intelli-

gence : a modern approach. Prentice Hall.

Sycara.K, Paolucci.M, Ankolekar.A, and Srinivasan.N

(2003). Automated discovery, interaction and com-

position of semantic web services. Journal of Web

Semantics, 1(1):27–46.

Tan.M, Nguyen.X, and Kowalczyk (2006). Modelling and

solving qos composition problem using fuzzy discsp.

In 2006 IEEE International Conference on Web Ser-

vices ICWS 2006.

Yuan.Y, Li.Z, and Sun.W (2006). Graph search based ap-

proach for BPEL4WS Test Generation. In IEEE In-

ternational Conference on Software Engineering Ad-

vances ICSEA’06.

WEBIST 2007 - International Conference on Web Information Systems and Technologies

296