Data as a Service (DaaS) builds on service-oriented technologies to enable fast access to data resources on the Web. Many approaches are proposed to achieve the DaaS composition task which is reduced to query rewriting problem. In this context, DaaS is described as Parametrized-RDF View (P RV ) over Domain Ontology (DO). However, the DO is unable to capture the di erent perspectives or viewpoints for the same domain knowledge. This limitation raises semantic con icts between pieces of data exchanged during the DaaS composition process. To face this issue, we present a context-driven approach that aims at supporting semantic mediation between composed DaaSs. The semantic reconciliation based on mediation service is performed through the execution of rule mapping which achieves the transformation between contexts.
Nowadays, modern enterprises are using Web services for data sharing within
and across the enterprise's boundaries. This type of Web service is known as
Data-as-a-Services (DaaSs) which return collections of Data for a given set of
parameters without any side e ects. DaaSs composition is a powerful means
to answer users' complex queries. Semantic-based approaches are proposed to
enable automatic composition by describing the Web services properties over
ontology. In fact, many ontology languages (e.g.,OWL-S3, WSMO 4) and
extension mechanisms (e.g., WSDL-S 5) provide standard means by which WSDL6
document can be related to semantic description. However, this means do not
provide a way to relate semantically the Web service parameters (i.e., input and
3 http://www.w3.org/Submission/OWL-S/
4 http://www.wsmo.org/TR/d2/v2.0
5 http://www.w3.org/Submission/2005/SUBM-WSDL-S-20051107/
6 Web Service Description Language
output) which hampers their applicability to DaaS composition. The
automation of DaaS composition requires the speci cation of the semantic relationships
between inputs and outputs parameters in a declarative way. This requirement
can be achieved by describing DaaS as views over a DO following the
mediatorbased approach [
Motivating example: Let us consider an e-health system where the
information needs of health actors are satis ed with DaaS Composition System
(DCS), as proposed by [
(Examen_id) S1
S2
S3 (BPR-state) a) b) (Patient_id) (Examen_id) (BPR.Value:BPR.D/BPR.S,mmHG) (BPR.Value : MAP,cmHG) (Cls.New) (BPR.Code) (LOINC→SNOMED) (BPR.Code)
MS1 S1
(BPR.Value) S3
MS2 MS3 (BPR.D/BPR.S→PAM) (mmHG→cmHG) (BPR-state) (BPR-state)
MS4 (Cls.New→Cls.Old) the appropriate mediation services, in the right order, to make the generated composition executable as depicted in gure 1.(b). The physician has to invoke: 1)\M S1": to mappe the BPR code returned by S2(LOINC9) to code acceptable by S3 (SNOMED10); 2) The composition of \M S2" and \M S3" where :\M S2" aggregates the two values expressing BPR measure returned by \S2" to M AP 11 value acceptable by S3 and \M S3" converts the M AP value expressed with the measurement unit (\mm/Hg") returned by M S2 to the M AP value expressed with the measurement unit acceptable by S3 (\cm/Hg"); \M S4" : to mappe the BP R state returned by \S3" represented according to the new classi cation BPR value table (e.g., stage 1,2,3,4) to the state acceptable by the physician represented according to the old classi cation (e.g., severe, moderate, mild). This is a rather demanding task for non expert users (e.g.physicians). Thus, automating con ict detection and resolution in DaaS composition is challenging.
Contributions: In this paper we propose a context driven approach for automatically inserting appropriate mediation services in DaaS compositions to carry out data conversion between interconnected DaaS. Speci cally, we propose 1) a context model expressed over Con icting Aspect Ontology(CAO) which is an extension of \DO"; 2) an extension of PRV based DaaS model based on context to express more accurately the DaaS parameters semantic; 3) a mediation service model behaving as a mapping rule to perform the transformation of DaaS parameters from one context to another.
Outline: The rest of this paper is organized as follows. Section 2, presents the overview of our approach. In Section 3, we leverage di erent proposed models. In Section 4, we present a global view on our con ict detection and resolution algorithm and our implementation. Finally, section 5 provides a conclusion and future works. 9 LOINC : Logical Observation Identi ers Names and Codes 10 SNOMED: Systematized Nomenclature of Medicine, Clinical Terms 11 Mean Arterial Pressure is BPR value, M AP = 32 (BP R:D) + 13 (BP R:S)
The DaaS composition process starts when the user speci es a query over
DO and CAO using SPARQL 12 query language (see circle 1 in gure 2). The
DCS uses the query rewriting algorithm proposed by [
We leverage in this section di erent models used through the paper. The de
nition of the basic concepts such as the Domain Ontology(DO), the Parametrized
RDF view (PRV) and the Conjunctive Query (CQ) are presented formally in
[
Con icting Aspect Ontology: Con icting Aspect Ontology (CAO) is a family of a lightweight ontology, speci ed in RDFS. CAO extends the DO entities with a taxonomic structure expressing di erent DaaS parameters semantic con ict13. The CAO is a 3 tuple < ACg; ACi; >, where: 1)\ACg" is a set of classes which represents the di erent con icting aspects of a DO entities. Each acg class in ACg has one super-class and a set of sub-classes. Each acg class has a name representing a con icting aspect, such as, CAO:Measurement-Unit as depicted in Figure 3; 2)\ACi" is a distinct set of instanceable classes having one super-class in ACg. By de nition, aci is not allowed to have sub-classes. For instance \mm=HG" and \cm=HG" are two instanceable classes from the CAO:BPR-Unit class; 3)\ " refers to the sibling relationships on ACi and ACg. The relationships among elements of ACg is disjoint. However, elements of ACi of a given acg can be related by the Peer relationship which indicates similar data semantics. Part-Of relationship which relates aci entity and its components (e.g., BPR.D and BPR.S values are Part-Of BPR).
CAO(BP_structure)
CAO: BPR.Value
Same as MAP
BPR.D/ BPR.S
Disjoint
CAO(system Code)
CAO: System-Code
Disjoint
CAO(Mesearment-Unit)
CAO:Mesearment-Unit
Disjoint Loinc. code
Same as Snomed. .code
ICD. code
CAO:BPR
Unit
Disjoint
CAO:Gaz.
Unit
Same as mmGH
cmHG Classe
Rdfs:SubClassof
Sibling relationship
13 For the classi cation of the various incompatibility problems in web service
composition see [
Context model: The context has the form: C = f(Di; Vi)ji 2 [1; n]g where Di, represents an acg class whose values are from a value-set (V i) where Vi 2 ACi. For instance, the context CMU = fBP R U nit : mm=HGg indicates that the BPR measurement unit is \mm/HG". The proposed context model is used to express more precisely the query formulated by the user, the DaaS published by the provider and the semantic con ict occurring in each O=I14 operation in given csK 2 CS. 1)\Contextualized Conjunctive Query model" is CCQ(X) : < CQ(X)jCCQ(X;CO) > where CQ(X) is the conjunctive query expressed over DO, and CCQ(X;CO) is the context of the distinguished variable X and the query constraint CO expressed over CAO; 2)\Contextualized DaaS model": The C-DaaS is Sj ($Xj ; ?Yj ) : < VDO > j < ExtCAO > where VDO is the PRV of Sj and ExtCAO is a tuple < CXj ; CYj > where CXj and CYj are respectively the input and the output DaaS parameter contexts. CXj and CYj are described by a set of RDF triples over CAO in form of 2-tuple < ACg; ACi >; 3)\Context and semantic con ict": In the sense of the present work, semantic con ict occurs in On=Im operation having respectively On and Im as an output and an input parameter which refer to the same DO entity. However, their contexts represented respectively by COn and CIm refer to di erent \aci" entities from the same \acg" . Then we say that a parameter semantic con ict "acg" exists in On=Im. 3.3
Mediation service model
Mediation Services M S assures the semantic reconciliation in the case where
the O=I operation causes a con ict. The M S model consists of mapping rule
having the form M S($OJ ; ?IJ ) : GO ! GI , where $OJ and ?IJ are the sets of
input and output variables of M Sj respectively. GO and GI represent the set
of RDF triples representing contextualized DaaS /query parameters. We deem
appropriate to use the SPARQL's construct statement (i.e., CONSTRUCT GI
WHERE GO) as a rules language to de ne rule mapping as proposed by [
DO:HasBprCode
BPR DO:CodeValue C Rdf:type
DO:HasBprCode
A Rdf:type
DO:CodeValue $x CO:HasBprCodetype CAO:BPC CO:HasBprCodetype ?y
CAO:LOINC CAO:SNOMED 14 i.e, two subsequent DaaSs \Sn" and \Sm" in \cs", First(cs) and CCQ(CO), Last(cs) and CCQ(X). each con icting aspect ACg we de ne a mapping rule template. For instance, the mediation service M S2 assuring the same-as mapping one-to-one of BP code value from \LOINC" code to \SNOMED" code is presented in gure 4. In the same manner, we de ne the mapping many-to-one, one-to-many and many to many. To the best of our knowledge, this work is the rst to use SPARQL construct statement to model mediation services. 4
In the following, we present the details of our Con ict Detection and Resolution Algorithm (CDR) depicted in gure 5. The inputs to the CDR is a set of
cs CS
cs CS-R cs R
Conflict Object Set {COS}
Mediation services repository
Composition «R» \CSs" generated by the QR algorithm as explained in section 2. The outputs of CDR are \CSs" con ict free. The desired mediation service is found and called automatically using the CDR algorithm which is two phases : Detection and Resolution. In the rst phase each composition \cs" is examined to detect potential con icts. Thus, if \cs" is without con icts then it is inserted into the set of compositions without con icts R; else the con icts of \cs" are added into the con ict object set \COS". Finally, the set of composition without con ict R is removed from CS. Thus CS consists of the composition with con icts. In the second phase, each detected con ict is resolved by performing the matching between the required context transformation to the mapping rules de ning the mediation services. The matching is obtained, the automatic calls to the correspondent mediation services are inserted in \cs" to resolve con icts. Then, the new set of composition CS (i.e, composition without con ict) are added into R and returned to DCS for query plan execution. In order to test test our proposal, we have implemented a Java Based application and test it with multiple examples, including the motivating example 15. Each Web services is deployed 15 The implementation test are available in http://sites.google.com/site/ehrdaas/home on top of a GlassFish web server. Each DaaS is annotated by the contextualize P RV and each Mediation service is annotated by SPARQL construct statement. In the evaluation phase we have considered a set of queries through which we identify the following : 1) During the detection phase, we can detect the set of con ict aspect identi ed in \ACg". 2) During the resolution phase, according to the number of con ict detected in each O=I operation: when there is a con ict including one aspect acg ( e.g., BPR-code) or a con ict including several aspects acg ( e.g., BPR-value), our solution provides automatically the appropriate mediation service. When we have a several mediation services allowing to resolve the same con ict, our algorithm returns randomly one of them as long as they achieve the same functionality. 5
In this paper, we propose an extension of PRV based DaaS model based on context. The proposed context model expressed over Con icting Aspect Ontology aims to handle semantic con ict in DaaS composition. Our model allows to specify the mediation service as mapping rule performing the simple or complex transformation of DaaS parameters from one context to another. Our future perspective will to deal with the performance issues of our algorithm and how to resolve a given con ict for which there is no appropriate mediation service.