WEB SERVICES-BASED QUERY REWRITING AND RESOLUTION

IN LOOSELY COUPLED INFORMATION SYSTEMS

Mahmoud Barhamgi, Pierre-Antoine Champin and Djamal Benslimane

LIRIS Laboratory, Claude Bernard Lyon1 University, 69622 Villeurbanne, France

Keywords:

Web services composition, Data integration, P2P systems.

Abstract:

Recently a great deal of information resources has been exposed in the form of Data-Providing Web services

in Peer-to-peer based e-collaboration environments. In this paper we model these services as RDF views over

the peer local ontology, then we use these views in establishing a composition of Web services that satisﬁes

a received query. Our composition is Data-Driven, therefore we apply extra data treatments on data ﬂow

between composed services before getting the desired result.

1 INTRODUCTION

In today’s loosely-coupled e-collaboration environ-

ments (e.g. eHealth, eGov...etc) the access to an in-

creasing number of data sources is made through Web

services. We call this kind of services as “Data-

Providing Services” as opposed to “Functionality-

Providing Services” since their invocation only re-

turns a piece of information without causing any

change in the environment (e.g. charging a credit

card, ...etc). Data-Providing services are very com-

mon in eHealth collaboration environments, in the

same health site, for example, they are used to en-

capsulate and integrate numerous proprietary data

sources that otherwise cannot be integrated, e.g. sen-

sors, equipments equipped with proprietary inter-

faces...etc. Across different sites, they are used to

share patients records, or as a means to transfer out-

sourced data. Such e-collaboration environments can

be modeled as peer-to-peer environments where every

peer holds a collection of DP services, puts them at

the disposal of its partners and, in return, they provide

it with their DP services. The collaboration implies

that some of the peer’s data items are outsourced or

stored at its partners and that it needs to exploit their

DP services to retrieve these items when needed.

So far, P2P Data Management and Integration

Systems (Halevy et al., 2004; Rodr

ıguez-Gianolli

et al., 2005; L

oser et al., 2003; Spyropoulou and Dala-

magas, 2006) were only concerned with handling tra-

ditional data sources (as opposed to services). In these

systems, peers are supposed to directly hold and ex-

pose their data, either in a syntactic form (XML),

or recently in some semantic forms (OWL instances

plus some inferencing capabilities), then when they

are interrogated, they apply queries squarely to data

instances in order to materialize answers. However,

none of these systems pay attention to the conse-

quences raised by the adoption of DP services for

data sharing. With this form of data accessibility, it

becomes impossible to materialize data in any forms

or structures before applying, in a subsequent step,

queries to it, rather the query resolution here neces-

sitates to decompose the received query in terms of

available services, compose these services, and to co-

ordinate their execution.

2 OUR APPROACH

In this paper we provide a condensed description

of our framework ((Barhamgi et al., 2007)) to pro-

vide better support for data sharing and integration in

eSystems that make an extensive use of DP services.

In our framework individual peers adopt the stack pre-

sented in ﬁgure 1 which has the following layers.

1. Data-Providing Services Layer. This layer holds

(or makes reference to) services that contribute data

items pertaining to the peer in question. These ser-

vices are either local or remote ones (i.e offered by

the peer’s partners), and they can be picked up based

on the SOA(Papazoglou, 2003) model.

2. Views Layer. In this layer we model previously

397

Barhamgi M., Champin P. and Benslimane D. (2007).

WEB SERVICES-BASED QUERY REWRITING AND RESOLUTION IN LOOSELY COUPLED INFORMATION SYSTEMS.

In Proceedings of the Third International Conference on Web Information Systems and Technologies - Internet Technology, pages 397-401

DOI: 10.5220/0001277303970401

 SciTePress

Figure 1: The Stack adopted by our peers for query resolu-

tion in eSystems adopting Data-Providing services.

selected services (in the ﬁrst layer) as RDF param-

eterized views over the peer’s local ontology. These

views serve us for query resolution in the next layer.

3. Data-Providing Services Composition and Execu-

tion Layer. Here we make use of previously deﬁned

views to decide what are the services whose composi-

tion can satisfy a received query (either a local query

or a query received from the peer’s acquaintances).

The rest of the paper is organized as follows. Next

we provide a running example. In section 4 we model

both our query and services. In section 5 we discuss

the issue of query rewriting then we conclude the pa-

per in section 6.

3 RUNNING EXAMPLE

Consider the case of the peer P1, it holds the ontol-

ogy in ﬁgure 2, and it has the following DP services.

Two remote DP services W S

and W S

to retrieve

Test A ,Test B (specializations of Test) respectively.

Two local DP services; W S

: it returns the medica-

tions list taken by a given patient. W S

: it returns

patients (their names) who have been administered a

given medication. Now assume the following query

on P1: “Q1: what are the tests performed by patients

who have been administered a medication termed as

“SomeStuff”?. Obviously the resolution of Q necessi-

tates the composition of several services, in particular

W S

, W S

and W S

(local and remote ones).

4 MODELING ISSUES

4.1 Our Queries

Given an ontology O(C, L, DP, OP), where C is the

classes set, L is the literals set, DP is the datatype

properties in O and OP is the object properties set in

O. Our queries have the following form.

{ { ?c

. Ψ . p

1.2

. ?c

. Ψ . p

2.3

.... Ψ . p

n−1.n

. ?c

}, CL, OS }, where:

1. ?c

...?c

are variables of types deﬁned by

classes within C,

2. p

i. j

is the object property linking ?c

to ?c

. Both

i:1→n

and p

i. j

constitute the query’s“backbone”.

3. Ψ is a linking operator and it is used when one

variable ?c

is linked to more than one other variable

such that each of these variables pose a condition on

the selection of ?c

4. CL is the constraints set imposed on datatype prop-

erties of ?c

i:1→n

5. OS is the output set, it comprises output variables

(and their projected datatype properties).

In the spirit of this deﬁnition our query became:

Q1: {?T1(Test) . [Has − Test]

−1

. ?P1(Patient) . [Take-Medication] .

?M1(Medication), Ct= {$M1(Name=“Some Stuff”)}, Out= {?T1(Result)}

}

4.2 Data-Providing Services as Views

Web services are usually modeled with the de facto

standard for service description OWL-S. In partic-

ular, OWL-S’s Service Proﬁle permits to model the

service’s functionality, inputs and outputs. However,

DP Services have no explicit functionality, instead,

the semantic relation holding between their inputs and

outputs must be captured. Therefore OWL-S may not

be the best choice for describing them since it does

not allow to capture this relation. We model Data-

Providing Services in our approach as RDF Parame-

terized Views (PVs) over OWL ontologies.

Each PV is a predicate W S

):- 4-tuple

<Backbone, Ct, In, Out> where,

W S

) is called the view head and it comprises

the name of corresponding service and its returned

results. The rest is called the view body and it has the

following contents:

1. Backbone it comprises both the variables set C (of

classes types) linking the input and the output of the

service, and the object properties set OP relating the

different variables in C.

2. Ct is the constraints set imposed on the datatype

properties of C without being required inputs of the

service.

3. In is the necessary literals for the service invoca-

tion.

4. Out is the output literals.

According to this deﬁnition, a view on one of our ser-

vices is showed in ﬁgure 3, the complete list of views

are presented in (Barhamgi et al., 2007).

WEBIST 2007 - International Conference on Web Information Systems and Technologies

398

Figure 2: An example of OWL ontology modeling the peer’s local data items and the items provided by its partners.

Figure 3: A Parameterized view deﬁned for WS1.

5 WEB SERVICE-BASED QUERY

REWRITING AND EXECUTION

Here we pay attention to the conditions under which

a composition is considered as valid, then we present

the pretreatments we apply on our views.

For formal discussion assume a query

Q(Q

backbone

, Ct

, OS

) and a set of services,

each has a PV

(Ser

i backbone

, Ct

, In

, Out

). In order to

satisfy Q, backbones union of selected services has

to cover the query’s backbone, the ﬁnal output of the

composition (the sum of Out

of selected services)

should satisfy OS

, and the Q’s constraints list Ct

satisﬁed with the union of Ct

. However some special

cases may occur while the query rewriting process;

we review them brieﬂy.

1). The union is larger than the query backbone with

provision to all of the asked outputs. Herein it should

be veriﬁed whether the additional concepts in the

union have a corresponding input parameter neces-

sary for the service invocation. Here the composition

cannot be invoked as a necessary input will not be

available and thus the composition is invalid.

2). OS

is not satisﬁed with the sum of Out

as some

literals do not appear in the output of the composition.

Herein the composition of these services is invalid.

3). A constraint speciﬁed in the Ct

was dropped

(e.g. patient gender). Herein if dropped constraints

were mandated then these services will be rejected.

4). The composition of the selected services enforce

an additional constraint that was not speciﬁed in the

query’s Ct

. Herein the composition is valid.

5). There is a conﬂicting constraint between Q

and one of the selected services (e.g. the gender

property has conﬂicting values male vs. female). The

composition herein is invalid.

6). The union of the services backbones does not

cover the query backbone. In this case these services

must be rejected.

All of these observations were dealt with in our Web

services-based query rewriting algorithm presented

in (Barhamgi et al., 2007).

5.1 Preprocessing the Deﬁned Rdf

Parameterized Views

Before the rewriting process, the parameterized views

should be preprocessed. This includes the following

steps.

Step 1. Extending the Obtained Pvs to Reﬂect

Owl “explicit” Subclassing Statements

We extend previously deﬁned PVs with the con-

straints subClassOf, subPropertyOf that are explicitly

declared in the ontology, e.g. in (ﬁgure 5, case A) a

new triple was added to the PV of W S

indicating that

an instance of “TestA” is also an instance of “Test”.

Figure 5: An extended PV for WS1. A new triple was

added (showed in bold)to reﬂect the relation between Test

and TestA.

WEB SERVICES-BASED QUERY REWRITING AND RESOLUTION IN LOOSELY COUPLED INFORMATION

SYSTEMS

399

Figure 4: The query rewriting process for the running example. There were two yielded rewritings for Q.

Step 2. Skolemizing Triples

Variables denoting classes in PVs need to be skolem-

ized (Chen et al., 2006), that is to replace each vari-

able by a skolem function helpful to merge instances

stemming from different services. An example of a

skolemized PV is shown in (ﬁgure 5, case B). The

properties of a skolem function for a particular class

are chosen by the domain expert.

Figure 4 shows the rewriting process of the query

in the running example.

5.2 The Rewritings Execution

Our algorithm for query rewriting yields a certain

number of DP services compositions. Before execut-

ing these compositions, we apply an extra algorithm

in order to superimpose these compositions (if possi-

ble). The intent here is to avoid the duplicate invoca-

tion of the same service across several compositions.

In our system the results of each service invocation

are materialized in the form of OWL instances before

being used to invoke subsequent services or sent to

the requester . We need to apply extra treatment and

processing over data ﬂow among combined services.

In general, three semantic operators can be applied.

They are as follows.

Semantic Union (W S

∪ W S

): This operator is used

to semantically combine the outputs (OWL instances)

of the services W S

, WS

. The outcome includes the

disjoint instances provided by W S

and W S

and the

semantically equivalent instances provided by both

only once.

Semantic Intersection (W S

∩ W S

): This operator

can be used to return the semantically equivalent in-

stances provided by both W S

and W S

Semantic Difference (W S

 W S

): It can be used

to return the instances provided by W S

excluding the

equivalent instances provided by W S

Note that treatments on data ﬂow are applied in the

execution time of the composition. Returning back

to our example, the execution of the compositions is

done as follows. First, W S

is invoked with medica-

tion name. The returned patients’ information is put

automatically in the form of OWL instances. Then,

for each obtained instance of patient we invoke both

W S

and W S

and the results are materialized as OWL

instances then sent to the requester.

Figure 6: Data ﬂow in Data driven Web services composi-

tion.

6 CONCLUSION

In this paper we have rapidly presented our frame-

work (Barhamgi et al., 2007) for query resolution in

terms of Data-Providing web services. We believe

that this work is of great importance since that data

most of the time is shared via web services in P2P

collaboration environments. Currently, we are focus-

ing taking into account the privacy constraints in our

framework.

REFERENCES

Barhamgi, M., Champin, P.-A., and Benslimane, D.

(2007). A Framework for Web Services-Based

WEBIST 2007 - International Conference on Web Information Systems and Technologies

400

Query Rewriting and Resolution in Loosely Cou-

pled Information Systems. Technical report,

http://liris.cnrs.fr/publis/?id=2647.

Chen, H., Wu, Z., Wang, H., and Mao, Y. (2006). Rdf/rdfs-

based relational database integration. In ICDE,

page 94.

Halevy, A. Y., Ives, Z. G., Madhavan, J., Mork, P., Suciu,

D., and Tatarinov, I. (2004). The piazza peer data

management system. IEEE Trans. Knowl. Data Eng.,

16(7):787–798.

oser, A., Siberski, W., Wolpers, M., and Nejdl, W. (2003).

Information integration in schema-based peer-to-peer

networks. In CAiSE, pages 258–272.

Papazoglou, M. P. (2003). Service-oriented computing:

Concepts, characteristics and directions. In WISE,

pages 3–12.

Rodr

ıguez-Gianolli, P., Garzetti, M., Jiang, L., Kementsi-

etsidis, A., Kiringa, I., Masud, M., Miller, R. J., and

Mylopoulos, J. (2005). Data sharing in the hyperion

peer database system. In VLDB, pages 1291–1294.

Spyropoulou, E. and Dalamagas, T. (2006). Sdqnet: Se-

mantic distributed querying in loosely coupled data

sources. In ADBIS, pages 55–70.

WEB SERVICES-BASED QUERY REWRITING AND RESOLUTION IN LOOSELY COUPLED INFORMATION

SYSTEMS

401