International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 63
Towards a Framework forAgent-Enabled SemanticWeb Service Composition
Vadim Ermolayev, Natalya Keberle, and Sergey PlaksinZaporozhye State University, Ukraine
Oleksandr Kononenko and Vagan TerziyanUniversity of Jyvaskyla, Finland
ABSTRACT
The article presents the framework for agent-enabled dynamic Web service composition. Thecore of the methodology is the new understanding of a Web service as an agent capabilityhaving proper ontological description. It is demonstrated how diverse Web services may becomposed and mediated by dynamic coalitions of software agents collaboratively performingtasks for service requestors. Middle Agent Layer is introduced to conduct service request totask transformation, agent-enabled cooperative task decomposition and performance. Dis-cussed are the formal means to arrange agents’ negotiation, to represent the semantic structureof the task-activity-service hierarchy and to assess fellow-agents’ capabilities and credibilityfactors. Finally, it is argued that the presented formal technique is applicable to variousapplication domains. Presented is the ongoing work on designing and implementing agent-based layered architecture for intelligent rational information and document retrieval. Fi-nally, the discussion of the OntoServ.Net framework for the development of P2P mobile serviceinfrastructures for industrial asset management provides the extension of the Web service com-position approach.
Keywords: asset management; composition; information retrieval; ontology; software agent;Web service.
INTRODUCTION
Web services are the emerging tech-nology promising to become one of thefuture key enablers of the Semantic Web.There are strong prerequisites that, beingself-described and self-contained modu-lar active components, Web services will
appear to be the key elements in assem-bling intelligent software infrastructures inthe near future.
There is the emerging consensus thatthe ultimate challenge is to make Web ser-vices automatically tradable and usable byartificial agents in their rational, proactiveinteroperation on the next generation of
64 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
the Web. It may be solved by creating ef-fective frameworks, standards and soft-ware for automatic Web service discov-ery, execution, composition, interoperationand monitoring (McIlraith et al., 2002).Personal opinion of the authors is that thelist should be extended by the means formaking services the subject of automatednegotiation and trade. It is also importantfor future service enabled Web infrastruc-tures to cope with business rules1, notionsand mechanisms of reputation and trustwith respect to services and service pro-viding agents, dynamic character, flexibil-ity, reconfigurability of partial plans(Ermolayev & Plaksin, 2002), workflows,and modeled business processes.
Current industry landscape providesonly initial and very partial solutions to theultimate problem. Existing de-facto stan-dards for Web service description(WSDL), publication, registration and dis-covery (UDDI), binding, invocation, andcommunication (SOAP) provide merelysyntactical capabilities and do not reallycope with service semantics. Known in-dustrial implementations, such as HP E-speak (Karp, 2003), base on these stan-dards and do not completely solve thechallenge of semantic serviceinteroperability. It should be mentionedthat major industrial players realize thenecessity of further targeted joint researchand development in the field (Layman,2003).
More recent research and standard-ization activities of DARPA DAML com-munity resulted in offering semantic ser-vice markup language DAML-S(Ankolekar et al., 2002) based on RDFplatform. The constellation of XML based
languages/ontologies for business process,logistics description is also expanding:WSFL, ebXML, BPML, RuleML,BPEL4WS …
The goal of the article is to highlightwhat should be still done on top of recentresearch accomplishments in order tomake Web services automatically tradableand usable by artificial agents in their ra-tional, proactive interoperation on the nextgeneration of the Web. Conceptual framesfor this development are under intensivediscussion and some proposals alreadyappear (e.g., WSMF (Fensel & Bussler,2002)).
The article offers a new understand-ing of a service as an intelligent agent ca-pability implemented as a self-containedsoftware component. From the other hand,provided that agents negotiate and tradeexchanging services in the process of theircooperative activities in open organiza-tions, a service may be considered (as,say, in E-speak) a kind of a generalizedresource. This approach evidently impliesthe appearance of the rational service pro-viding agent demanding certain incentivesand aiming to increase its utility. If, for ex-ample, a service requested from a travelagency is ‘BookRoundtrip(‘Kiev’,‘Erfurt’, 22/09/2003, 25/09/2003, …)’,the price paid by the requestor will com-prise the prices of consumable (DAML-S, 2003) resources (air fare, hotel room,…) plus the incentive paid to the serviceholder for ‘BookRoundtrip’ service com-ponent usage. This remark seems to berational as far as we pay either the salaryto the office manager or a fee to a travelagent, who make arrangements for us in ahuman-business environment. Moreover,
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 65
it is not in the eye of the service requestor,but the agent performing ‘BookRoundtrip’service will realize according to the ser-vice markup (or the Partial Local Plan(PLP) in our terminology (Ermolayev etal., 2001)) that the requested process(DAML-S, 2003) (or the task in our ter-minology (Ermolayev et al., 2001)) iscomposite and will require cooperationwith at least air companies’ service pro-viding agents and hotel booking serviceproviding agents. These independent ac-tors will evidently also intend to increasetheir own utilities by requesting fees fortheir services.
The article first provides the over-view of the basic notions, approaches andarchitectural solutions with respect to agentparadigm, WWW and the Semantic Web,Semantic Web enabled Web services.Detailed discussion of the popular travelplanning scenario helps to claim that full-scale Web service exploitation needs so-lutions beyond the facilities of today’s se-mantic service markup. The article focuseson one of the major open problems —dynamic composition of a desired com-plex service by a coalition of rational co-operative freelance agents.
Next it is argued that it is a reason-able architectural solution to introduce anAgent Middle Layer (e.g., Sycara, et al.(1999)) between services and serviceconsumers. Negotiation on Web serviceallocation based on the authors’ approach(Ermolayev & Plaksin, 2002) is proposedas the mechanism for dynamic compositeservice formation. DAML-S (DAML-S,2003), our Task and Negotiation On-tologies (Ermolayev et al., 2001) are usedfor service dynamic composition and to
facilitate inter-agent-operability.Further on it is described how the
approach to dynamic agent-based servicecomposition is applied to intelligent ratio-nal information retrieval from distributedautonomous resources. Finally, theOntoServ.Net (Kaykova et al., 2004;Terziyan, 2004; Terziyan & Kononenko,2003) framework and the aspects of ser-vice mobility and service adaptation arediscussed. The architectural principles forservice composition in a peer-to-peer ser-vice network are also outlined.
WHAT AN AGENT IS
Agent paradigm in software engi-neering is one of the powerful means tonarrow the semantic gap between theconceptualizations we use to analyze andto model the phenomena of the real worldand the resulting distributed software sys-tem. If compared to the objects in OOSE,which may be interpreted as the analogyof inanimate entities in the real world,agents generally represent animate objects,typically human beings. Intelligent softwareagents are therefore used when the soft-ware needs to possess some “human” fea-tures like the ability to perceive the envi-ronment and reactivity, apparent proac-tive behavior in succeeding at a goal onbehalf of the human owner, ability to learnfrom their experience, and social behav-ior. One of the inherent intelligent featuresof agents is the ability to form social struc-tures — teams, communities, coalitions,and organizations. A rational agent asthe member of a social structure needs tobalance its individual rationality andbenevolence in facilitating to the growth
66 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
of the group utility. Agents often use ne-gotiation mechanisms adopted from hu-man encounters for that. An agent alsoneeds to obey its social commitments andthe conventions that regulate the groupbehavior within the social structure. A teamor an organization of agents that cooper-ate in a physically and, possibly, geographi-cally distributed network form a softwaresystem called a Multi-Agent System(MAS). An agent and an MAS are themain conceptual patterns of the AgentOriented Software Engineering (AOSE).
From the engineering perspective, atthe lower level of abstraction, the essen-tial features of agents in MAS are theirabilities to communicate with each otherand to coordinate their activities. Coordi-nation means achieving coherence in thegroup activities and thus providing that thesolution of a problem or the accomplish-ment of a task is obtained with less effort,fewer resources consumed, and betterquality. Communication stands for the abil-ity to exchange the pieces of informationwithin an encounter in a uniform way andusing shared terminology. Communicationamong agents in open systems, which aretypical in the majority of real- world casesin e-business, enterprise application inte-gration, and so forth, is a challenginginteroperability task. The solutions areapproached by standardizing the commu-nicative languages (e.g., FIPA ACL) anddeveloping formal machine-processablerepresentations of the common terminol-ogy in the form of ontologies. Ontologies,formalized in ontology description lan-guages (e.g., OWL) provide: aconceptualization — a formal model of thereal world phenomena in a domain; a vo-
cabulary — a set of terms or symbols iden-tifying concepts; and an axiomatization —the rules and the constraints on conceptsand their properties that capture charac-teristic aspects of the domain.
Agent paradigm and AOSE gainmore and more popularity as one of thekey enablers of the emerging SemanticWeb — the new generation of the Webwhose abstract architecture is outlined inW3C WWW TAG Architecture Specifi-cation.
More details may be borrowed from,for example, Ermolayev and Plaksin(2002) and Jennings (2000).
W3C WWWARCHITECTURE
WWW Architecture provides theabstract specification of the architectureof the Web. It figures out the conceptualmodel, the properties and the semanticsof the WWW constituents, and defines theunderlying principles and the basic con-straints of Web-based system develop-ment. WWW architecture specificationfixes the design choices approved by W3Cand approves the good practices of usingthe Web technology that guide futuregrowth, and consistent and successfulevolution of the Web.
The primary task of W3C Techni-cal Architecture Group (TAG) is todevelop and maintain the consensualspecification of the basic principles of theWeb technology in order to facilitate andcoordinate cross-technology architecturedevelopments inside and outside W3C.TAG claims identification, interaction,
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 67
to the capabilities and the heterogeneityof Web agents involved.
W3C WEB SERVICESARCHITECTURE & THESEMANTIC WEB
Web Service Architecture speci-fies generic concepts and defines theframework for the creation of Web ser-vices. Web services are modular softwarecomponents accessible over a WWW. AWeb service is supplied with the descrip-tion specifying its interface in a machine-processable way to provide for theinteroperability in open distributed soft-ware systems. The description contains thespecification of the message formats,datatypes, and transport and serializationprotocols.
The following de-facto industrialstandards outline today’s technologicalframes for Web service development andpublication: WSDL — Web Service De-scription Language, UDDI for UniversalDescription, Discovery and Integration,SOAP (Simple Object Access Protocol)for Web service binding and invocation,and XML and HTTP for serialization.However, ongoing research activities moveforward the state-of-the-art by develop-ing extensible ontology-based frameworkfor the Semantic Web enabled Web ser-vices.
W3C Semantic Web Initiativeaims primarily to provide a comprehen-sible framework for identifying, represent-ing and processing the semantics of Webresources. The ultimate vision of the Se-mantic Web is the worldwide distributed
and representation as the key aspectsof Web architecture and derives its ab-stract specification from these concepts.
Identification on the Web is basedon the semantics and the use of the URIs(Uniform Resource Identifiers), which areglobal identifiers and are central to theWeb architecture.
Interaction is defined by TAG asthe communication of resources that in-volves URIs, messages, and data amongagents over WWW. TAG provides thebasic concepts for messages, Web agents,interaction styles, and the use of metadataand the protocols for agents. TAG alsodefines the architectural constraints andthe assumptions for agent interaction andthe patterns for human-user interaction onthe WWW.
Representation of data on theWeb is grounded on the defined conceptsof media types, data formats, encoding,namespaces, general hypertext infrastruc-ture and the use of XML as the core lan-guage. It is worth mentioning in the con-text of the representation aspect that therepresentation of metadata on the Webis not explicitly defined by the Web archi-tecture specification yet and is likely to bebased on the Semantic Web principles forthe next generation of the Web.
The Separation of Content, Pre-sentation, and Interaction is yet onemore of the most important principles ofthe Web architecture. It concerns the de-velopment of the standards for highlyinteroperable distributed systems in openand dynamic environments, where infor-mation is created, accessed and processedat the high level of autonomy with respect
68 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
device for computation, inhabited withartificial service providing agents. It istherefore extremely important to haveWeb service semantics formally and ex-plicitly represented in a machine-processable way. Such semantic repre-sentations in the form of ontologies areessential for automated service discovery,invocation, orchestration and trade andevidently extend the current technologicalframes. Semantic Web resources and ser-vices will have semantic annotations —small ontologies providing both a meta-description of the resource and the vo-cabulary of the relevant concepts. Seman-tic Web initiatives spend substantial efforton ontology language (RDFS, DAML,OWL) development and standardization.
DAMLS & SEMANTICWEB ENABLED WEBSERVICES
The concept of Semantic Web en-abled Web Services (SWWS) is the syn-ergy of Web service technologies with thesemantic Web framework. It assumes thatthe semantic Web infrastructure is the toplayer of the conventional WWW. This se-mantic layer contains Web service ontolo-gies, notations and standards for servicedescription, facilities for service discov-ery, orchestration and integration. SWWSwill be widely used in the future Web,where intelligent agents will discover Webservice providers, reason about their ca-pabilities by analyzing their semantic de-scriptions and dynamically compose ser-vices on demand through cooperation withthe service, providing agents having ap-propriate capabilities.
One of the pioneering targetedSWWS initiatives is the development ofDAML-S (DAML-based Web serviceontology). DAML-S is the extension ofDAML+OIL ontology language. It speci-fies the core set of concepts for describ-ing the granularity, the properties, the ca-pabilities and the grounding of a Web ser-vice. If compared to current industry stan-dards, DAML-S provides a higher de-gree of flexibility and expressiveness indescribing service semantics, allows mod-eling of extensible service hierarchies andtype systems, and provides the means tospecify the constraints and the rules forWeb services.
TRAVEL PLANNINGSCENARIO
Let’s consider the mentioned travelplanning scenario having in mind that ourintentions have become true and Web ser-vices are available at the desired level ofsemantic interoperation. The authors haveplayed the following exercise assumingthemselves as “intelligent software agents”participating in cooperative execution ofa conference trip planning task (Figure 1).Each agent possessed his/her beliefs aboutthe environment and capabilities in per-forming one or another activity related tothe overall high-level goal achievement —‘BookRoundtrip(“Kiev, Ukraine”, “Erfurt,Germany”, 22/09/2003, 25/09/2003,“ICWS’03-Europe”, …)’. Agents’ capa-bilities were: their knowledge of relevantWeb sites providing human-oriented ser-vices and their ability to operate these ser-vices via Web interfaces. Agent roles were:
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 69
• AUTHOR (A) — an agent representingone of the article’s authors intending toattend ICWS’03-Europe and request-ing ‘BookRoundtrip’ service
• TRAVEL AGENT (T) — an agent actu-ally providing ‘BookRoundtrip’ serviceby generating and conducting corre-sponding task execution
• FARE AGENT (F) — agents providingvarious airfare information and book-ing services
• ICWS INFO (I) — an agent providinginformation services on ICWS’03-Eu-rope local arrangements, infrastructure,accommodation, and so forth in Erfurt
• HOTEL AGENT (H) — agents providinghotel room reservation services
• BUSINESS PARTNER (P) — an agentrepresenting A’s business partner inAustria with whom A intends to meet inGermany at the time of the conferenceto discuss a joint proposal
As usual, in travel planning an A iscapable of just invoking a T with‘BookRoundtrip’ task, to formulate his orher constraints, preferences and needs forspecial arrangements, and to approve so-lutions proposed by the T. According to‘BookRoundtrip’ description in terms oftask ontology (Ermolayev et al., 2001)known both to A and T (but with differentgranularity), service inputs are2:Starting_Point= “Kiev, Ukraine”Destination=“Erfurt, Germany”Beg_Date =22/09/2003End_Date=25/09/2003Event=“ICWS’03-Europe”Preferences=(“low fare”, “fast connections”, “4-star hotel”, “continental breakfast”, “conference discounts”)Constraints =(Budget = •1500, Payment=(VISA, USD), Hotel >= 3-star,Room-per-night <= •110,Hotel_Location=”in Max 20 min walk fromthe Conference venue”)Special_Arrangements=( ( Event=“businessdinner”,
Agent = (“Prof. Heinrich C. Mayr”, http://www.ifi.uni-klu.ac.at/ IWAS/HM/Staff/Heinrich.Mayr/ ), Date=(23/09/2003-24/09/2003), Location=(Erfurt, Munich)),
The process starts with the arrange-ment (Ermolayev & Plaksin, 2002). Aundertakes to hire one of the Ts as thecontractor for the job. This arrangementis performed in the frame of the ExtendedIterative Contract Net negotiation as de-scribed further in Section 3.4. The flow ofround trip booking, which T performs forA, is presented on Figure 1. At first T ac-cepts the task from A by means of agents’communication interface. This interfacemay be built upon ACL (FIPA, 2003) forFIPA3-compliant agents (e-Appendix A-14). T then uses its beliefs on how to‘BookRoundtrip’(e-Appendix A-2), for-malized according to the task ontology (e-Appendix A-6), to derive that the acceptedtask is complex and involves at least‘PlanTrip’, ‘MakeHotelRes’, ‘ApplyForVisa’, ‘SpecArrangements’ and ‘ApproveSolution’ activities. ‘PlanTrip’ activity ischosen (PLP of task ontology (Ermolayevet al., 2001)) as the first to be performed
and appears to be also a complex task:‘InquireFares’, ‘ApplyConstraints’, …,‘BookFare’, ‘ApproveSolution’. Beforeoutsourcing fare inquiry to F, T ‘notices’that a slight change in the starting or end-ing date of the trip may result in a sub-stantial decrease in the airfare expensesbecause of the Sunday Rule discounts5
commonly offered by air companies.For our example this means to T that
the dates 20/09-25/09 and 22/09-28/09should be also rationally considered forthe trip. T negotiates these input changeswith A, asking A to provide desirabilityvalues for these dates (Figure 2 — graydots) indicating max price A is ready topay for the fare within the specified dates.Requirements, which T specifies for‘InquireFares’ service, are thus slightlychanged by introducing the list of datepairs for which the service should be per-formed. Contract Net negotiation is theninitiated by T having Fs as participants.
F-s propositions,6 resulting from‘InquireFares’ service execution, are alsooutlined on Figure 2. These results causethe necessity to use one more service,which was not initially planned by T’s PLPfor the task. As far as the offers are pro-
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 71
vided in different currencies, T needs tochange the task and require the servicefor currency conversion7 (+(‘Convert Cur-rencies’, e-Appendix A-3), Figure 1).Conversion results are presented on Fig-ure 2. It is now easy for T to derive thatthe acceptable proposition is still for thedates 22/09-25/09, but with the destina-tion at Frankfurt (not at Erfurt), whichwere not initial ‘BookRoundtrip’ task in-puts from A. However, this result com-plies with A’s preferences as far as thereare nonstop flights available from Kiev toFrankfurt (but not to Erfurt and Munich).This implies the necessity for T to‘AdjustPreferences’ by inquiring A’s ser-vice. The mechanism may be similar toinputs negotiation discussed above and theoutcomes may cause the invocation ofsome new activities; for example, changeto a train at Frankfurt-Main Airport —inquire the ‘BookRailwayFare’ service fromDie Bahn8 Agent. Discussion of theseemerging task branches is omitted, as faras it is conceptually similar to that alreadygiven before. It is, however, important tonotice that activities that were not initiallyplanned often emerge and appear to becritical to the overall goal achievement notonly in the discussed scenario.
It is not informative to discuss sub-sequent activities of T. Hotel booking andvisa application services are performedmerely in the same manner and agents usesimilar mechanisms of task decompositionand negotiation for that. Special arrange-ments list is also considered as the list oftrip planning tasks. However, it should bementioned that the execution of these ac-tivities should be properly coordinated:note for instance that hotel reservation re-
quires that the fare has been alreadybooked as precondition (check-in andcheck-out dates, money left) and GermanConsular Service may require that the fareand the hotel room have been bookedbefore issuing the visa.
Other important aspects, not men-tioned before, are the ones of credibility,trust and meaning negotiation amongagents participating in cooperative taskperformance and service composition.Recall special arrangements input for theillustration. T will negotiate with P on vari-ous aspects while arranging the businessdinner. The dilemma for P in this environ-ment is if to trust T (as the contractor of A,which is the trusted one because of thelong record of partnership) and allow himor her to make the arrangements for P, orto reason that A may be not really experi-enced in arranging business dinners inGermany and to decide to better rely onhis or her credible (Section 3.4) partnersfrom Germany. In the latter case P will in-form T that it will better arrange the eventon its own. This in turn may affect the ne-cessity of the approval from A.
COOPERATIVE DYNAMICSERVICE COMPOSITION
Let’s enumerate the features neededto rationally provide composite flexibleservices for the automation of the sce-narios, like that of travel planning, in an e-business environment.
Intelligent service provider needs tobe capable of:• Understanding the semantics of the ac-
tivity it is supposed to perform, reason-ing on if the activity is atomic or com-
72 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
Figure 3: Agent-Based Service Provision Mediation Framework
plex, decomposing complex activitiesaccording to its knowledge and the ex-perience of the environment;
• Adjusting activity inputs, requestor pref-erences and constraints in order toproactively reach the higher-level goal
• Negotiating with the requestor and theother service providers in a rational wayon optimal service performance, allo-cation in order to increase its own util-ity or to obtain common meaning of theservice inputs, outputs, preconditionsand after-effects;
• Monitoring and assessing credibility andtrustworthiness of other service provid-ers to minimize risks;
• Coordinating services performanceflow according to the inputs and pre-conditions.
It seems obvious that service pro-viding distributed open software systemspossessing these capabilities may be mostnaturally designed and assembled of soft-
ware agents. Agent platforms and agent-based systems are already used for ser-vice brokerage (McIlraith et al., 2002),matchmaking (Sycara et al., 1999), andcoordination (Papadopoulos, 2001). Theremainder of this section will shortlypresent the formal approach to dynamictask decomposition and performance bycoalitions of rational agents (Ermolayev etal., 2001; Ermolayev & Plaksin, 2002).
Middle Agents forService Composition
Conceptual idea of service media-tion is not originally new and has been ar-gued by many authors. Strong mediationhas been, for instance, claimed as one ofthe basic principles for WSMF (Fensel &Bussler, 2002). However, the frameworkfor intelligent dynamic service composi-tion according to the changes in the envi-ronment affected by the service executionflow has not been worked out before.
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 73
The proposal of the MediationFramework for Agent-enabled ServiceProvision targeted to dynamic servicecomposition is presented on Figure 3.Control flows are labeled with legends initalic; data flows are marked by bold leg-ends. The principles on which the proposalis centered are:• Agent-based middle layer is required
for scalable, intelligent, dynamic servicecomposition;
• Composite services are interpreted astasks comprising activities of varyinggranularity by the agent middle layer;
• Service mediator is formed dynamicallyas the coalition of service providingagents (SPAs) participating in the taskexecution;
• SPAs join task coalitions only for thetime their service is required for the re-spective task;
• SPAs are economically rational(Nwana, 1996), autonomous and in-dependent in making their decisions —the only fact one SPA believes aboutthe behavior of another SPA is: it is in-dividual and rational (Sandholm, 1996).
• SPAs are capable of: incoming task de-composition according to their localknowledge (task ontology, PLP); mak-ing arrangements for activity outsourcingto other SPAs based on extended it-erative contract net negotiation; activ-ity outsourcing to the chosen contrac-tor SPA; adjusting their beliefs on otherSPAs’ capabilities and evaluating SPAs’credibility through monitoring coopera-tive activities;
• Services are self-contained modularloosely coupled program componentswrapped by SPAs; an SPA may allow
another SPA to use its service by pro-viding service context relocation;
• Specialization of an SPA is defined bythe set of services it wraps.
If the framework is examined fromthe point of implementability with existingservice markup solutions the state of af-fairs may look like given on Figure 3. Yetunsolved or partially unsolved problemsof service mediation are:• Lack of common semantic ground and
commonly accepted mechanism foractivity outsourcing, activity parametersadjustment and meaning negotiation —negotiation ontologies family;
• Lack of common specifications/crite-ria for capability monitoring, credibilityand trustworthiness assessment.
The proposed architectural layeringis likely to remain valid for the request-task-activity-service ontology hierarchy: aservice request is interpreted as the taskat the requestor layer; these tasks are de-composed into activities at the middlelayer; and activity descriptions actuallywrap service markups. The remainder ofthe section provides some outlines to ap-proach the solutions of the open issues.
Request-Task-Activity-ServiceHierarchy
The semantic hierarchy for a request-task-activity-service reflects the principlesof the proposed architectural layering. A
74 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
request belongs to the sphere of ServiceRequestor Layer and is specified in termsof task ontology (Ermolayev et al., 2001).The function of the SPA chosen as thecontractor for the specified request is todetermine if the incoming task is the atomicactivity according to its local specifications(task ontology). In case the task is com-plex and should be decomposed intoatomic activities at the local level of granu-larity, the next round of activities alloca-tion negotiations is initiated. Only the ac-tivities the given SPA is not capable toperform on its own are negotiated withother SPAs, while the ones correspond-ing to the initiator’s capabilities are sched-uled for self-performance. Only an activ-ity for which it is true that: (a) it is atomicand (b) the SPA is able to perform it on itsown, is in the relationship with the corre-sponding service or service loop. Atomicactivity execution is performed by the SPAthrough invoking its capability macro-model (Ermolayev & Plaksin, 2002): ac-tivity context is translated into DAML-Smarkup corresponding to service profile;the service is then invoked via the inter-face specified by its binding (or groundingin terms of DAML-S) description. Ser-
vice invocation loop may actually result inone or several service runs depending onthe wrapping activity inputs. For example,‘InquireFares’ service will be performedthree times as far as three different dateintervals are to be processed (Figure 2).
Semantic facet of request-task-ac-tivity-service layering is presented on Fig-ure 4. Specifications for ‘InquireFares’activity and service are given in e-Appen-dix A-5.
Capability & Credibility Assessment
SRA and SPAs are to be able todetermine which of the SPAs are capableto perform the task to be allocated. Pos-sible mechanism to define the perspectivecontractors is capability matchmaking(e.g., based on LARKS (Sycara et al.,2002)), or service discovery techniquebased on UDDI, or another service match-ing facilities (e.g., semantic matching basedon DAML-S profiles (Paolucci et al.,2002)). However, in case there is somecapability beliefs record maintained au-tonomously by an SPA in the course ofcooperative task execution, the use of thisknowledge may substantially facilitate low-ering computation costs by eliminatingunnecessary directory/matching serviceusage. Evidently, if A believes that B, Cand D are capable of performing desiredactivity because they did it before, it willrather proceed to contracting negotiationwith B, C and D directly instead of tryingto find some other SPAs9 with matchingcapabilities.
A model and a mechanism of agents’capability assessment based on SPA be-liefs representation in the form of Fellows’
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 75
Capability Expectations Matrix (FCEM)has been elaborated in frame of the re-ported research (Ermolayev & Plaksin,2002). SPAs accumulate and adjust theirlocal beliefs on the capabilities of theircollaborators from the experience of co-operative performance. New portions ofthis knowledge appear each time an ac-tivity is being outsourced to an SPA. Sub-jective beliefs of the SRA on the prob-abilities of its fellows’ capabilities to per-form the given activity are thus updated.FCEM for capability beliefs representa-tion is maintained in the following form:
1
11 1 1 1
1
... ...
SPA
...
... ... ( , ) ...
...
SPA
j m
j k
j j ji i i
j mn n n n
a a a
c c c
c q p
c c c
= =C (1)
where dimensions m and n change reflect-ing the appearance of new incoming ac-tivities and newly discovered or perishingSPAs.
Capability estimations cji change
each time an agent negotiates with its fel-lows on outsourcing an activity. Elementqj
i in tuple cj
i stands for the quantity of
recorded negotiations with fellow agent
SPAi concerning activity ja . Element pj
i
stands for the capability expectation. Therule for cj
i updates is as follows:
1.2
.1
+←
+←
ji
ji
ji
ji
ji
qqq
rpp
, (2)
where r is equal to: 0 — if the fellow re-jected the activity, 0.5 — if the fellow re-
plied that it can accept the activity and 1— if the activity was finally allocated tothe fellow.
One more aspect providing influenceon a task requestor’s decision to allocatean activity to one or another negotiationparticipant is its assessment of theparticipant’s credibility. A self-interestedSPA, due to the appearance of the newhighly attractive activity offers in the com-petitive environment or due to the pecu-liarity of its behavior, may lower previouslydeclared capacity (Ermolayev et al., 2001;Ermolayev & Plaksin, 2002) it is spend-ing for the bulk of the activities under ex-ecution. This will lead to the increase ofthe performance duration, which maytherefore seriously decrease therequestor’s desirability of these resultsand, thus, lower the credibility value forthe SPA selling its fellows short.
Let, for example, a serviceoutsourced to an SPA be ‘DeliverAir Tick-ets’. The result of the service is: the tick-ets are at the gate counter. The agreeddelivery time is 30 minutes before thecheck-in, though the deadline advertisedby the SRA before is the time when thecheck-in starts. The SRA will evidentlyconsider the SPA that delivered the tick-ets before or right in the agreed time ascredible. However, if the SPA delivers thetickets in five minutes before the check-in, the SRA may rightfully feel aggrieved,though it still has the chance to check infor the flight. The credibility of the SPA inthe eye of the SRA will therefore be low-ered. Further on, if the tickets appear atthe counter after the check-in has beenopened already, the SRA may rightfullyconsider that the contract terms were se-
76 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
riously violated by the SPA. Its credibilityshould be therefore drastically lowered.Finally, imagine an SRA still waiting for itstickets at the counter when the plane isalready taking off. In the latter case theSRA may even want to require a penaltyin addition to lowering SPA’s credibilityto zero. To summarize, it is natural to mea-sure the changes of an SRA’s beliefs onthe SPA’s credibilities by the losses of thedesirability of the service results based onthe stricken contract deal (refer to Fig-ure 5).
The mechanism of accounting fel-lows’ credibility values is similar to that ofadjusting the beliefs on changing fellows’capabilities (1-2). Credibility assessmentvalues change over time as the requestoragent adjusts its subjective beliefs by com-paring the desirability values (Figure 5)derived from:• 1st – activity duration the executive
committed to within the activity alloca-tion arrangement negotiation and
• 2nd – actual results delivery time. Cor-responding credibility matrix elementsare then recomputed due to the follow-ing:
>≤<
≤×=
ar
araraa
ar
jiji
dtdttttp
ttCrCr
,0),/(
,1: ,, , (3)
where: ta is the time the parties have
agreed to accomplish the activity a, tr is
the actual time of a results delivery, da is
the deadline and pa is the weight coeffi-
cient characterizing the current priority ofa for the activity requestor agent.
Credibility threshold values associ-ated with respective activities and storedin agents’ PLPs are used by task request-ing agents to assess possible risks and al-ter their strategies.
Negotiation on Activity Allocation
As it was mentioned above, nego-tiation on activity allocation takes place
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 77
each time an agent realizes, according toits knowledge of the activity or becauseof the overload, that the activity should beoutsourced to one of the fellow SPAs. Anextension of the FIPA Iterated ContractNet protocol has been proposed as theinteraction protocol for this kind of nego-tiation (see Figure 6). A service requestoragent is considered an Initiator (I) in thisencounter. The SPAs about which I be-lieves that they are capable to perform theactivity (FCEM) form the party of the in-vited Participants (P).
The first round of the interaction,which is actually the extension of the FIPAprotocol, aims to find out if any of theknown capable Ps may agree to performthe activity. Negotiation set for this roundcontains activity signature only (for ex-ample, ‘DeliverAirTickets’). An I may startexploring other opportunities of
outsourcing the activity if all Ps from thesphere of its awareness (Ermolayev &Plaksin, 2002) refuse in the first round.For example, I may require the list ofmatching SPAs from the MatchmakerAgent (MA, see Figure 8).
Negotiation on the second and thesubsequent rounds is about the terms ofthe possible contract. An I advertises theactivity inputs and the discrete results de-sirability function as the incentive overtime. I than chooses the best Ps proposalweighted by the respective credibility val-ues in case several Ps proposals result inthe agreement. Subsequent rounds areused to adjust the activity inputs or thedesirability function in the case if no oneof the Ps has agreed on the previous round(for example, dates, destination point onFigure 2).
Ps refusals and propositions are
Figure 6: Extended Iterated FIPA Contract Net Protocol
I ��P
final round
non- final
reject
adjusted CfP
reject
accept (agreement)
Initial CfP: service signature
refuse CfP
deadline
CfP deadline
activity deadline
inform
failure
Rou
nd 1
R
ound
2 +
sub
sequ
ent
propose
CfP: service inputs+desirability
refuse (disagree)
propose (agree)
78 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
shown on Figure 7. These feedbacks areformulated in a constructive way to allowthe I to adjust its CfP in the subsequentround. A feedback contains two incentive-time points defining the segment on whicha possible agreement may be stricken.Evidently, the area of agreement for thecurrent round could be formally definedas the union of all those parts of the feed-back segments that are on and below the
I’s desirability function polyline. All otherpoints of Ps’ feedbacks indicate their dis-agreement with the offer of the currentnegotiation round.
An I considers the negotiation roundas final if it can accept one of the Ps’ agree-ment and strike the contract deal. Thechosen P thus becomes the contractor andcommits itself to the task coalition for thetime necessary to perform the outsourcedactivity. Task coalitions are considered tobe a kind of social structure. Coalitionmembers are thus bounded with coalitioncommitments and convention regulatingtheir ratios of self-interest and benevolence(Ermolayev et al., 2001).
Negotiation ontology (Ermolayevet al., 2001) is used as the namespace andthe formal semantic frame for the contentsof the messages agents communicate withwhile negotiating on activity allocation.
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 79
RACING10
FUNCTIONALITIES,AGENTS, & SERVICES
A reader might argue that, fairly,travel planning is not the task that reallyrequires sophisticated agent-enabled au-tomation technique: negotiations, coali-tions, service wrapping and composition— at least from the customer’s side. Travelplanning is not that time consuming to makeits performance impossible without auto-mation. Moreover, a human will some-times still be better in arranging looselyformalized things that require intuition andcontext- dependent understanding withcomplexity beyond the capacity of, say,the first order logic based languages. How-ever, the presented technique is applicablenot only in case you plan your conferencetrip (Ermolayev et al., 2001; Ermolayev& Tolok, 2002).
Let’s project the above discussionto distributed information and documentretrieval domain. In the terms of documentretrieval a service request is commonlyformulated as a search phrase — a firstorder logic expression over the list of key-words or phrases. Documents (Webpages, scientific papers, magazines,books) are stored at disparately structureddistributed autonomously maintained da-tabases or text collections in a digital form,are marked-up according to different stan-dards and often cost money. A task fordocument retrieval may thus be presentedas the set of interrelated activities distrib-uted over the document providers. Theseactivities wrap the (partial) queries derivedfrom the initial user’s request.
The goal of the RACING project isto provide mediation facilities for userquery processing by the means of thequery semantic decomposition, the ratio-nal distribution among independent, au-tonomous, rational document retrieval ser-vice providers wrapping respective docu-ment resources, and the fusion of the ob-tained results (Figure 8). User agents act-ing on behalf of the human users or realorganizations (e.g., libraries) and serviceproviding agents are considered as busi-ness representatives or business modelsin frame of the project. RACING media-tion may thus be classified as B2B media-tion. It is evident that such a kind of intel-ligent activities really needs sophisticatedautomation to be scalable and gracefullydowngradable.
User query processing, resourcewrappers registration by the capabilitymatchmaker and common ontology main-tenance are the basic functionalities of theRACING mediator (Figure 8). Thoughonly query processing may be consideredas a real business process involving third-party service providers for money, theother two ones are also performed as tasksand require various types of negotiationand semantic interoperation.
For example, the outline for the userquery processing scenario is as follows.The process starts at UA with the formu-lation of the query in terms of the keyphrases familiar to the given user. UAs arecloned by CLA utility agent each time anew user comes to the mediator and per-ish when the user leaves. User profiles(mappings of their most frequently usedkey words or phrases to the MediatorCommon Ontology (MCO) concepts) are
80 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
incrementally collected, stored at OA(Ermolayev et al., 2003) in the form ofthe reference ontology and used by QTAs.UA actually generates and conducts thetask of query processing and acts as theproxy between the user and mediator.Query processing task generated by UAcontains ‘CloneQTA’, ‘TransformQry’,‘CloneQPA’, ‘ExecuteQry’ activities. Thecloning activities are outsourced to CLA,which clones QTA and QPA for queryprocessing. ‘TransformQry’ activity isoutsourced to QTA, which performs thetransformation of the query in terms ofkeywords to semantically matching queryin terms of the concepts of the MCO. Thelast activity is outsourced to QPA, whichgenerates the following set of activities for‘ExecuteQry’ task: ‘DecomposeQry’,‘PerformQryset’. Query decomposition isperformed by QPA in order to extract theparts of the incoming query that may re-quire different capabilities from documentservice providers. This extraction is guidedby topic classification of the MCO. Re-sulting set of partial queries is performedby QPA as the following activity sequence:‘MatchRWA’, ‘PerformQry’. Matching ac-tivity is allocated to MA for a certain in-centive over accomplishment time. MAreturns11 the list of RWAs capable to per-form document providing services relevantto the partial query. ‘PerformQry’ activityallocation is negotiated with pre-selectedRWAs in terms of service ‘overheads’ overtime and document price and the contrac-tor is chosen for query performance (Sec-tion 3.2). Contractor RWA receives thepartial query in terms of MCO. It there-fore needs to transform the query into theterms of its resource ontology. This trans-
formation activity is outsourced to OA,which actually holds the necessary map-pings. RWA than invokes document ser-vice that it wraps with the transformedquery and provides documents relevantto the query to QPA.
SERVICE COMPOSITIONIN P2P SERVICENETWORKS
One of the essential prerequisites forthe implementation of a RACING-likeservice composition platform is the provi-sion of the proper underlying infrastruc-ture. It becomes even more important inthe cases when the environment requiresmore sophisticated capabilities than thoseprovided by the conventional WWW. Thissection presents the OntoServ.Net frame-work (Terziyan, 2004; Terziyan &Kononenko, 2003) for the intelligent com-position of Web services on the SemanticWeb enabled industrial environment.OntoServ.Net is the agent-enabled frame-work for the management of industrialdevices in the peer-to-peer network ofmaintenance Web services. InOntoServ.Net the principles of the Se-mantic Web are used for the developmentof interoperable Web services and ontol-ogy-based information management. Peer-to-peer technology provides the means toorganize the communication infrastructure,and agent technology enables the imple-mentation of the problem-oriented behav-ior of network components (Terziyan,2003).
OntoServ.Net is a fully decentralizedenvironment that is a peer-to-peer net-work comprising service platforms located
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 81
at maintenance sites and service provid-ing centers. P2P structure of OntoServ.Net reflects existing approaches towardsthe creation of business-partnership envi-ronments where companies can share re-sources (in particular, Web services) thatwere previously used only internally. En-largement of such resource sharing envi-ronments heads towards a global P2Pnetwork with highly independent nodes.Though semistructured architecture willlikely be used (with large service centerswithin newly created communities), peer-to-peer interactions reflect the reality oftoday’s businesses.
Maintenance of complex industrialmachinery, for example a paper mill, re-quires hundreds of factors to control andinvolves many services to monitor varioussensor data, analyze general conditionparameters, performance, and so forth.Hardware configuration varies from onemachine to another, and thus, requires anindividual approach to the organization ofthe maintenance process and servicing.
The set of condition monitoring andmaintaining services in OntoServ.Net isdynamically composed depending on thecurrent needs of a machine. It changeswhen a fault state processing is required,or some service is substituted by the otherone in order to provide more efficiency orto follow degradation processes along themachine’s lifetime. OntoServ.Net servicenetwork improves performance and main-tenance quality by providing the most ap-propriate services available on the net-work.
Recently the synergetic approachesto the design of service infrastructurescombining the features adopted from the
Semantic Web, Web services and P2Pcomputing are under intensive research.Latest results prove the great potential ofsuch combinations for cooperative use ofdistributed heterogeneous informationsources and services (see e.g., Terziyan,2003; Sivashanmugan et al., 2002). Ser-vice discovery and composition of Seman-tic Web-enabled Web services in a de-centralized network present new chal-lenges for research community and de-mand thorough study.
In addition to a P2P structure of theservice network, OntoServ.Net presentsnew aspects related to the service com-position problem which were not thor-oughly studied before: service mobility,individual rationality of SPAs and their in-tended readiness to cooperatively workin a P2P environment.
Service Mobility
The specificity of the maintenanceactivities performed by the services inOntoServ.Net requires that these servicesare mobile. The reasons are: a need forguaranteed service availability, a need forminimization of the communication trafficover the network during long-term ser-vicing due to costs and/or technical re-strictions, strict constraints for service re-sponse time, security and privacy issues,and so forth.
Service mobility may naturally beimplemented if the services are providedby mobile agents able to migrate betweenagent platforms. Mobile services persiston the local service platforms on the siteand terminate after servicing. Actually, ser-vice instances arrive to a local platform
82 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
and are withdrawn later. However, somedata may be returned to the original SPAto update its knowledge base regardingthe performed diagnostics and efficiencyof actions taken. This knowledge is usedlater on for the improvement of the ser-vice quality (Terziyan, 2004).
Rational Agent-Services
OntoServ.Net services are wrappedby SPAs. SPAs, in addition to providingtheir services on SRAs’ requests, reasonabout which activities to perform in a givencase. OntoServ.Net has no division forservice requestor and service provider lay-ers, since both services and agents areconceptually the same. Resource Wrap-ping Agents (RWAs) represent industrialmachines or their parts and provide Webservices to grant access to or operationon the respective devices. RWAs also actas SRAs. For example, they acquire ad-vanced diagnostic services from anotherSPA to monitor basic parameters of themachine.
Resource wrapping agent shell(OntoShell, a framework for resource andservice adaptation to the Semantic Web-enabled environment (Terziyan, 2003))can be applied for a wide range of re-source types, including humans, knowl-edge bases and industrial devices.OntoShell allows wrapping services imple-mented within the framework of W3CWeb service architecture or, in principle,any other software development technol-ogy that provides external application pro-gramming interfaces.
Service Composition Strategy inOntoServ.Net
Service composition in OntoServ.Net is performed by platform-manageragents that act as mediators between ser-vice agents scattered over the network andlocal RWAs. A platform manager controlsservices’ mobility and supports the P2Pdiscovery mechanism of the OntoServ.Netenvironment, which is based on thematchmaking of a service request to dy-namic service profiles (Kaykova et al.,2004; Khriyenko et al., 2004). A profilepresents not only the service interface andthe semantics, but also comprises the gen-eralized description of SPA’s successful-ness in some states of the previously ser-viced SRAs. A dynamic profile is there-fore required for credibility assessment(Section 3.3). Since services are assumedto implement various learning techniques,their quality highly depends on the previ-ous invocations, the samples for self-learn-ing collected by SPAs, and initial trainingsets.
If a service is complex (Section 3)and requires the invocation of other ser-vices, the performance is conducted by alocal platform manager. The platform man-ager agent performs service discovery ei-ther locally or network-wide and providesinter-platform communication facilities.
To round up, the OntoServ.Netframework provides the means for thedevelopment of agent-enabled P2P Webservice infrastructures in the networks ofcomplex industrial machinery. The frame-work is applied to the development of thebusiness models and the implementation
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 83
of the secure service platforms that sup-port new type of mobile services. It isbased on the synergy of P2P and the Se-mantic Web, which ensures the success-ful deployment of industry-strong solutionsbased on agent technology.
CONCLUDING REMARKS
The paper presented the frameworkfor agent-enabled dynamic Web servicecomposition. The core of the methodol-ogy is the new understanding of a Webservice as an agent capability havingproper ontological description. It is dem-onstrated by the example of the travelplanning how diverse Web services maybe composed and mediated by dynamiccoalitions of software agentscollaboratively performing tasks for ser-vice requestors. It is also claimed that sucha mediation facility may substantially en-hance today’s solutions available in Webservice provision. This vision is groundedon the results obtained in agent-enabledbusiness process modeling and manage-ment.
It is stated that though the conceptof service mediation is not totally new thereis still some work to be done before itbecomes a real engineering technology.For example, the framework for intelligentdynamic service composition and decom-position according to the changes in theenvironment affected by the service ex-ecution flow has not been explicitlyworked out before. The framework in-troduces the agent middle layer to con-duct the transformation of a service re-quest to the corresponding task and forfurther cooperative task decomposition
and performance. Outlined are the formalmeans to arrange agents’ negotiation onactivity allocation, to represent the semanticstructure of the request-task-activity-ser-vice hierarchy and to assess fellow-agents’capabilities and credibility factors. Furtheron, it is argued that the presented formaltechnique is applicable not only to thetasks like travel planning. Presented is thereference architecture of the rational multi-agent mediator for intelligent informationand document retrieval. Further develop-ment and deployment of the mediator is inprogress in the frame of the RACINGproject. Presented aspects of service com-position and mobile-agent service repre-sentation in a peer-to-peer network ofservice integration platforms extend RAC-ING principles of service composition bythe aspects of mobility. The experience ofapplying OntoServ.Net framework to thedevelopment of P2P service infrastructuresprovides also the evidence of the applica-bility of the agent-enabled Web servicecomposition framework to real-world in-dustrial applications
Though thorough standardization andharmonization work should be performedbefore the presented approach becomesan engine for Web service provision, theauthors are certain that agent-enabled ra-tional Web service composition and me-diation may provide a substantial contri-bution, bringing closer the day when thebrave new world of machine-processableautomated Web services comes true.
ENDNOTES
1 International Workshop on RuleMarkup Languages for Business Rules
84 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
on the Semantic Web, 14 June 2002,Sardinia (Italy) http://tmitwww.tm.tue.nl/staff/gwagner/RuleML-BR-SW.html. Diffuse: Guide to Web Ser-vices http://www.diffuse.org/WebServices.html
2 Service inputs are given semi-formallyin order to avoid unnecessary detailsand save space.
3 Foundation for Intelligent PhysicalAgents, http://www.fipa.org/, last ac-cessed on April 24, 2003.
4 e-Appendixes A-1 – A-7 may bedownloaded from http://eva.zsu.zp.ua/services/app.htm
5 “One of the most common low fare re-strictions is the requirement for your stayto incorporate at least one Sunday. Forexample, for a round-trip New York toMiami, a passenger flying Tuesday toThursday might pay £328, but a pas-senger whose stay includes a Sundaywould pay much less - £188.” – http://www.flightcatchers.com/helpmenu/Howtofindcheapestfare.htmlast accessed on April 24, 2003.
6 Lufthansa Infoflyway Booking Servicehttp://lufthansa.com/ (last accessed onJuly 15, 2003) and Cyber Flyer Book-ing Service http://cyberflyer.galileo.com/(last accessed on July, 15, 2003) wereused in the described exercise to ob-tain the offers from F-s.
7 CNN Currency Converter: http://qs.money.cnn.com/tq/currconv/ lastaccessed on July 16, 2003.
8 http://www.bahn.de/, last accessed onJuly 16, 2003.
9 Applying to a capability registry maystill appear to be necessary in case B,C and D fail to provide constructive
proposals.10 RACING: Rational Agent Coalitions
for Intelligent Mediation of InformationRetrieval on the Net. http://www.zsu.zp.ua/racing/ Project fundedby the Ukrainian Ministry of Educationand Science under the grant No0102Y005339.
11 As QPAs in RACING have limitedlifetime, RWAs’ credibility and capa-bility assessment (Section 3.4.) is per-formed by MA for registered resourcewrappers. QPAs supply MA with nec-essary data obtained from cooperationwith RWAs.
REFERENCES
Ankolekar, A., Urstein, M., Hobbs,J.R., Lassila, O., Martin, D., McDermott,D., McIlraith, S.A., Narayanan, S.,Paolucci, M., Payne, T., & Sycara, K.(2002, June 9-12). DAML-S: Web ser-vice description for the Semantic Web.Proceedings of Int. Semantic Web Con-ference, Sardinia, Italy.
DAML-S. (2003). The DAML ser-vices coalition. DAML-S: Semanticmarkup for Web services. Retrieved June3, 2003, from http://www.daml.org/ser-vices/daml-s/2001/10/daml-s.pdf
Ermolayev, V., Eberle, N., & Tolok,V. (2001, November 27-30). OIL ontolo-gies for collaborative task performance incoalitions of self-interested actors. In H.Arisawa, Y. Kambayashi, V. Kumar, H.C.Mayr & I. Hunt (Eds.), Conceptual Mod-eling for New Information SystemsTechnologies ER 2001 Workshops,HUMACS, DASWIS, ECOMO, andDAMA, Yokohama, Japan.
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 85
Ermolayev, V., Keberle, N., Plaksin,S., & Vladimirov, V. (2003). Capturingsemantics from search phrases: Incremen-tal user personification and ontology-drivenquery transformation. In M. Godlevsky,S. Liddle & H. Mayr (Eds.), Informationsystems technology and its applications.Proc. of the 2nd Int. Conf. ISTA2003Lecture Notes in Informatics (LNI), P-30, GI-Edition, Bonn, 9–20.
Ermolayev, V.A., & Plaksin, S.L.(2002). Cooperation layers in agent-en-abled business process management.Problems of Programming, 1-2, 354-368.
Ermolayev, V.A., & Tolok, V.A.(2002). Modelling distant learning activi-ties by agent task coalitions. In Q. Jin, J.Li, J. Cheng, C. Yu & S. Noguchi (Eds.),Enabling society with information tech-nology. Tokyo: Springer-Verlag.
Fensel, D., & Bussler, C., (2002).The Web Service Modeling FrameworkWSMF. White paper. Retrieved June 3,2002, from http://www.cs.vu.nl/~dieter/wese/wsmf.paper.pdf
Jennings, N.R. (2000). On agent-based software In K.P. Sycara, M.Klusch, S. Widoff & J. Lu. (1999). Dy-namic service matchmaking among agentsin open information environments.SIGMOD Record, 28(1), 47-53.
Karp, A. (2003). E-speak e-xplained. HPL-2000-101 20000807External Technical Report. Retrieved June3, 2003, from http://www.hpl.hp.com/
Terziyan, V., & Zharko, A. (2004). For-mation scenarios in OntoServ.Net - Glo-bal network of intelligent industrial main-tenance Web services. IASTED Interna-tional Conference on Databases andApplications, DBA-2004, (to appear).
Khriyenko, O., Kononenko, O., &Terziyan, V. (2004). OntoEnvironment:An integration infrastructure for distributedheterogeneous resources. IASTED Inter-national Conference on Parallel andDistributed Computing and Networks,PDCN-2004 (to appear).
Layman, A. (2001, April 11-12).Web services framework. Proceedings ofW3C Workshop on Web Services 2001,San Jose, CA, USA. Retrieved June 3,2003, from http://www.w3.org/2001/03/wsws-program/
McIlraith, S.A., Son, T.C., & Zeng,H. (2002). Semantic Web services. IEEEIntelligent Systems, Special Issue onThe Semantic Web, 16, 46–53.
Paolucci, M., Kawamura, T., Payne,T., & Sycara, K. (2002, June 9-12). Se-mantic matching of Web services capa-bilities. Proc. of Int. Semantic Web Con-ference (ISWC’2002), Sardinia, Italy.
Papadopoulos, G.A. (2001). Mod-els and technologies for the coordinationof Internet agents: A survey. In A. Omicini,F. Zambonelli, M. Klusch & R. Tolksdorf(Eds.), Coordination for Internet agents- Models, technologies, and applica-tions. New York: Springer-Verlag.
Sandholm, T. (1996). Negotiation
86 International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004
among self-interested computationallylimited agents. PhD thesis, University ofMassachusetts, Amherst.
Siebes, R. (2002, October 30).Peer-to-peer solutions in the SemanticWeb context: An overview. EU-ISTProject IST-2001-34103 SWAP. VrijeUniversiteit Amsterdam.
Simple Object Access Protocol(SOAP) 1.1. W3C Note (2000, May 8).Retrieved June 3, 2003, from http://www.w3.org/TR/2000/NOTE-SOAP-20000508/
Sivashanmugan, K., Verma, K.,Mulye, R., & Zhong, Z. (2002). SpeedR:Semantic peer-to-peer environment fordiverse Web service registries. FinalProject, CSCI 8350. Enterprise Integra-tion.
Sycara, K., Widoff, S., Klusch, M.,& Lu, J. (2002). LARKS: Dynamicmatchmaking among heterogeneous soft-ware agents in cyberspace. AutonomousAgents and Multi-Agent Systems, 5,173–203.
Terziyan, V. (2003). Semantic Webservices for smart devices in a “global un-derstanding environment”. In R.Meersman & Z. Tari (eds.), On the moveto meaningful Internet systems. LectureNotes in Computer Science, 2889, 279-291.
Terziyan, V. (2004, March). Seman-tic Web services for smart devices basedon mobile agents. Fourth InternationalICSC Symposium on Engineering In-telligent Systems (EIS-2004), (to ap-pear).
Terziyan, V., & Kononenko, O.(2003). Semantic Web enabled Web ser-vices: State-of-art and industrial chal-lenges. In M. Jeckle & L.-J. Zhang (Eds.),Web Services - ICWS-Europe 2003,Lecture Notes in Computer Science,2853, 183-197. Springer-Verlag.
UDDI Technical White Paper. Re-trieved June 3, 2003, from http://uddi.org/
Revised Papers, LNCS, 2465,390-402. New York: Springer-Verlag,Berlin Heidelberg.
Thaden, U., Siberski, W., & Nejdl,W. (2003). A Semantic Web based peer-to-peer service registry network. Tech-nical report.
Web Services Description Language(WSDL) 1.1. W3C Note 15. (2001,March). Retrieved June 3, 2003, fromhttp://www.w3.org/TR/2001/NOTE-wsdl-20010315/
International Journal of Web Services Research, 1(3), 63-87, July-Sept 2004 87
Vadim Ermolayev is an associate professor at the Department of Information Technologies ofZaporozhye State University and the lead for the Intelligent Systems Research Group. Hisresearch interests include intelligent information integration, agent-enabled e-business and e-learning, ontologies, semantic Web services, evolution and adaptability in intelligent softwaresystems, and dynamic business process management and performance. He received his PhD inMathematical Modeling and Computer Science from Zaporozhye State University. More infor-mation may be retrieved from http://google.com/search?q=Ermolayev
Natalya Keberle is an assistant professor at the Department of Information Technologies ofZaporozhye State University and the member of the Intelligent Systems Research Group. Herresearch interests include the ontological models and the methods for presenting evolvingsemantics in intelligent information systems. More information may be retrieved from http://google.com/search?q=Keberle
Oleksandr Kononenko is a PhD student at the Department of Mathematical Information Tech-nology of the University of Jyväskylä and a member of Industrial Ontologies Group. His re-search concerns semantic-enabled industrial asset management and resource adapters forSemantic Web-based environments. Contact him at [email protected].
Sergey Plaksin is a PhD student at the Department of Information Technologies of ZaporozhyeState University and a member of the Intelligent Systems Research Group. The focus of hisresearch is within coordination patterns and models for agent-enabled business process man-agement and service provision. Contact him at [email protected].
Vagan Terziyan is an associate professor and a senior researcher at the Department of Math-ematical Information Technology of the University of Jyväskylä. He is also the chair of theDepartment of Artificial Intelligence and Information Systems at the Kharkiv National Univer-sity of Radioelectronics. His research interests and competences comprise distributed AI,multiagent systems, Semantic Web and Web services, peer-to-peer computing, knowledge man-agement and machine learning. He received a Dr.Tech in Engineering degree from KharkivNational University of Radioelectronics. More information may be retrieved from http://google.com/search?q=Terziyan
