An Approach to Semantic Matching of Web Services

Transcription

1 An Approch to emntic Mtching of Web ervices Dingjin Chen Jin Wu huyu Li Mnfu M Zhengguo Hu Computer school orthwestern Polytechnicl University Xi n hnxi 7007 Chin Computer school hnxi orml University Xi n hnxi 7006 Chin Abstrct ne of the chllenging problems tht Web service technology fces is the bility to effectively discover services bsed on their cpbilities An pproch is proposed to tckle this problem in the context of description logics The semntic informtion of Web service cn be described with three fctors: functionl cpbilities ction nd non-functionl ttributes We formlize these three fctors mtchmking nd propose novel mtchmking lgorithm tht tkes s input service request nd n ontology T of services nd finds set of services whose functionl cpbilities ction nd non-functionl ttributes mtching rnks re ll highest We then present simple exmple to illustrte our pproch to semntic mtching of Web service Keywords: Web services Discovery emntic mtchmking Description logics 0 Introduction The emntic Web services vision is to describe Web services cpbilities nd content in n unmbiguous computer-interpretble lnguge nd improve the qulity nd robustness of existing tsks such s Web service discovery nd invoction emntic Web services will lso enble brod rnge of new utomtion tsks tht humns previously performed including utomted composition interopertion execution monitoring nd recovery To support this vision emntic Web services will provide more powerful Web service development tools tht enble mong other things utomted simultion nd verifiction of Web service properties nd consistency-checking nd debugging fetures Exmples of such efforts include DAML- [] WMF [] nd METER- Work in this re is still in its infncy Mny of the objectives of the emntic Web services prdigm such s description of service cpbilities dynmic service discovery nd gol driven composition of Web services hve yet to be reched ur work focuses on the issue of dynmic discovery of Web services bsed on their cpbilities Dynmic service discovery is usully bsed on the rtionle tht services re selected t run-time bsed on their functionl cpbilities ctions nd non-functionl ttributes The key problem of Dynmic service discovery is the mtchmking of Web services ur im is to ground the discovery process on mtchmking between requester query nd vilble Web service descriptions We formlize the service mtching pproch bsed on semntic in the context of description logics (DLs) [] A key spect of DLs is their forml semntics nd resoning support DLs provide n effective resoning prdigm for defining nd understnding the structure nd semntics of concept description ontologies This is essentil for providing forml foundtions for the envisioned emntic Web prdigm [4] Indeed DLs hve hevily influenced the development of some emntic Web ontology lnguges (eg DAML+IL or WL [5]) We dopt WL- to describe the emntic Web services 0 Principles for mtchmking We suppose tht services nd requests re expressed in DL L equipped with model-theoretic semntics We suppose lso tht common ontology for services nd requests is estblished s TBox T in L ow mtch between services nd requests R could be evluted ccording to T According to WL- Web service is constructed with service profile process model nd grounding [6] o we give now forml definition of Web service s definition Requests hve the sme forms with services Definition : (Web service) A Web service is tuple ( x xn) ( F P ) where: - is the nme of the service - x x n re individul vribles - F describes the functionl cpbilities of the service In more detil F is tuple F = ( I ) where I nd re the inputs nd outputs of the service nd re the preconditions nd effects of the service All the four elements re expressed with DL formul - P describes the ction of the service In more detil P is tuple P = ( Pr Ef ) where Pr is

2 the set of preconditions nd Ef is the set of postconditions (Referring to DDL [7]) - describes the non-functionl ttributes of the service nd is expressed with DL formul Consequently in order to check whether the service mtch with the request R t first we should check mtching reltions between F nd F R P nd P R nd R respectively then colligte these three mtching reltions to get the mtching reltion between with R This frmework ensures the first rule tht we would like to hold for mtchmking nmely n open-world ssumption Rule (pen-world descriptions): The bsence of chrcteristic in the description of proposl should not be interpreted s constrint of bsence Insted it should be considered s chrcteristic tht could be either refined lter or left open if it is irrelevnt for the issuer of the proposl Moreover if ll constrints of request R were fulfilled by service but not vice vers then should be mong the top rnked supplies in the list of potentil prtners of the requester while R should not pper t the top in the list of potentil prtners of the supplier Rule (on-symmetric evlution): A mtchmking system my give different evlutions to the mtch between service nd request R depending on whether it is trying to mtch with R or R with ie depending on who is going to use this evlution Bsed on these two rules we give some lgorithms for mtchmking of the three fctors ie functionl cpbilities ction nd non-functionl ttributes From high to low we lso define four clss mtching rnks for every fctor which re Exct trong Wek nd ot- Mtch bviously the service which hs highest mtching rnk of ll fctors is the best But in generic cses consumer would like more interesting of functionl cpbilities thn the others or they would so much s hve no interesting in ction or non-functionl ttributes so we dopt the mtchmking principle like this: (Mtchmking principle): Let C be the set of ll cndidte services First evlute the mtching rnks of functionl cpbilities of ll cndidte service in C Then remrk ll services which hve highest rnk nd remove the rest which re not mrked from C If consumer hs interest in mtching ction or non-functionl ttributes then evlute the mtching rnks of ction or non-functionl ttributes of ll service in C At lst choose the service which hs highest rnk s the most pproprite mtchmking The reminder of this pper is orgnized s follows ection describes the formliztion of semntic mtching of Web service in the context of DL-bsed ontologies ection 4 presents n exmple of the proposed service mtchmking technique We review relted work in ect 5 nd provide concluding remrks in ect 6 0 Mtching pproch Functionl cpbilities mtchmking In this section we ll first provide the pproch to the functionl cpbilities mtchmking of Web service As mentioned in definition the semntic of functionl cpbilities of Web service is described by IPE ie F = ( I ) o the functionl cpbilities mtchmking cn be reduce to respectively check the subsumption reltions of IPE between service nd request First some definitions re presented s follows Definition : Let f : A { Δ be mpping from ssertions to the set of concepts nd roles () { C if ϕ = C() { R if ϕ = R() b f( ϕ) = f( π) f( ψ) if ϕis n ssertion which is composed with ssertion π nd ψ () f ( ϕ ) is the number of concepts in f ( ϕ ) C f ( ϕ ) is the number of roles in f ( ϕ ) nd R f( ϕ) = f( ϕ) C + f( ϕ) R Definition : (Reduced et of concept) Let L be DL T be set of xioms in L () A = { A A n is clled reduced cluse set if either n = or no cluse subsumes the conjunction of the other cluses: i n: Ai A \ A i The set A is then clled reduced set form (RF) of every description C A () The RF of description C A cn be constructed s follows: Let C = { A A n be the set of ll elementry concepts of C For ny Ai Aj i j n if let C = C A i until C is RF the set C is then clled generl reduced set form (GRF) of C If contrrily let C = C A j until C is RF the set C is then clled specific reduced set form (RF) of C Definition 4: (tructurl subsumption nd tructurl equivlent) Let { A A n nd { B B m be the RFs of description A A nd description B B Bm () A nd B re structurl equivlent (denoted by A B ) iff: ( n = m) ( i n j k n: Ai Bj Bi = Ak) () A nd B re structurl subsumption (denoted by A B ) iff: ( n = m) ( i n j n: A B ) Definition 5: (the semntic of functionl cpbilities of Web service): Let ( F P ) be Web service described with WL- F = ( I ) describes the semntic of functionl cpbilities of : i j

3 () I I In nd m re clled input nd output expression of I = { I I i j n nd = { i j m i j i j re the GRF of input nd output () ϕ ϕn nd ψ ψ m re clled precondition nd effect expression of f ( ) = f ( ϕ ) f ( ϕ ) nd n f( ) = f( ψ) f( ψm) re hold Intuitively the RF hs more semntic constrint thn GRF Hence in order to improve recll on the premise tht precision is little influenced we weken the semntic constrints of the service nd request both ie using GRF of input nd output to describe the semntic constrint Then we lso define the mtchmking of concept s definition 6 nd the mtchmking of ssertion is defined s definition 7 Definition 6: (the mtchmking of concept) In the semntic mtching of web services let A be the concept which is trget B be the concept which is cndidte (or source) ie checking whether B cn mtch A () B nd A re strong mtchmking (denoted by M ( BA) ) iff: B A B A () B nd A re wek mtchmking (denoted by M ( ) W BA) iff: B A trong mtchmking nd wek mtchmking re lso ll clled pproximte mtchmking denoted by M ( BA ) () B nd A re not mtchmking (denoted by M ( BA) ) iff: B A A B Proposition : In the semntic mtching of web services the mtching reltion between two concepts is nonsymmetric Proof Referring to rule nd definition 6 it is nturlly proved Definition 7: (the mtchmking of ssertion) In the semntic mtching of web services let ϕ be the ssertion which is trget ψ be the ssertion which is cndidte (or source) ie checking whether cn ψ mtch ϕ ψ nd ϕ re mtchmking (denoted by M A( ψ ϕ )) iff: ( f( ψ) C= f( ϕ) C) ( f( ψ) R= f( ϕ) R) ( α f( ϕ) β f( ψ) β α if α nd β re concepts or β α if α nd β re roles) therwise ψ nd ϕ re not mtchmking Hence we cn define the functionl cpbilities mtchmking Definition 8: (the functionl cpbilities mtchmking) In the semntic mtching of web services let ( F P ) be Web service nd F ( I ) ( F P ) be request nd F ( I ) ε be the concept miss error () hs Exct functionl cpbilities mtching reltion with iff: ( A I B I M ( B A) ) ( A B M( BA )) ( I = I ) ( = ) MA( M A( ) () hs trong functionl cpbilities mtching reltion with iff: ( A I B I M ( B A) ) ( A B M( B (( I = I ) ( = ) MA( ) M A( ) ) () hs Wek functionl cpbilities mtching reltion with if one of the following conditions comes to be true: ( A I B I M( B ( A I B I MW ( B ( A B M ( B ( A I B I M( B ( A B M( B ( A B MW ( B ( A I B I M( B ( A I B I MW ( B ( B M( B ( A B M ( B A W 4 { ( { A A B M( B A) ε) ( A A I B I M ( B A) ε ) (4) If it is out of ll the bove situtions hs ot- Mtch functionl cpbilities mtching reltion with The mtching directions of inputs nd outputs re contrry For the inputs mtchmking it is better to mtch request s inputs with service s inputs ie whether the request provides sufficient input prmeters to the service n the contrry it s better to mtch service s outputs with request outputs nmely whether the service cn produce pproprite outputs required by the request ince the preconditions nd effects of the service hve little effect on functionl semntics for simplicity in the definition of wek mtching reltion of functionl cpbilities the mtchmking of precondition nd effects hve been omitted Tht is there exists wek mtch reltion between service nd request if the wek mtch reltion of the input/output semntics hs been stisfied in spite of the mtch of precondition nd effects The Concept Miss Error refers to the mount of missing concepts in the input semntics or in the output semntics between the request nd the service This prmeter could be ssigned by users or by service mtching engine ccording to the empiricl vlue In the second pproch the service mtching engine could mke some recommendtory conclusions (listing those services whose concept miss error is within given scope) when it could not find ny service tht closely mtching the request The serch recll could be djusted by the scope prmeter obviously the concept miss error in inverse proportion to the recll Thus through the DL resoning mchine the semntics-bsed web service functionl cpbilities mtchmking procedure is composed of two steps: () executing the input/output semntics mtching lgorithm between the service nd the request () if the result of step stisfies the strong mtch nd precondition nd effect requirements exists in the request executing the precondition nd effect mtching sub-procedure Besides )

4 the orgniztion of the WL- bse is hierrchicl structure which clssifies the WL- ccording to the domin txology And the Domin prmeter defines the specil domin tht the discovery request interests As result mtching engine tht occupies more precise domin ontology bse could only serch WL- belonged to the sme domin which improves the mtching efficiency nd recll In the following lgorithm we denote how to implement the functionl cpbilities mtchmking Algorithm FunctionMtchmking( ε Domin); input: request ( F P ) nd F ( I ) concept miss error ε the correltive Domin of request the set of cndidte services T = { n output: set of services whose function cpbilities re mtchmking with request function RFMtchmking( BM TM ) input: trget concept description BM source concept description TM output: trongmtch(-) WekMtch(- ) otmtchumber Action mtchmking Action mtchmking is defined s follows Definition 9: (the ction mtchmking) In the semntic mtching of web services let ( F P ) be Web service ( F P ) be request () hs Exct ction mtching reltion with iff: P P ie P P P P () hs trong ction mtching reltion with iff: P P P P () hs Wek ction mtching reltion with iff: P P P P (4) If it is out of ll the bove situtions hs ot- Mtch ction mtching reltion with Therefore ction mtchmking of Web service depends on subsumption reltion between ctions of service nd request We stte the following method tht checks the subsumption reltion between ctions wrt ABox nd TBox Theorem : Let α = ( Pα Eα ) nd β = ( Pβ Eβ ) be ctions There re not ny individul vribles in α nd β We cll α β iff: () For A P β if A is constnt condition (no vribles) A should fulfill one of these two cses t lest: (i) B P α B is constnt condition nd A B (ii) B P α B is vrible condition nd A unifies with B ie there is permuttion θ such tht A = B θ therwise if A is vrible condition (hs vribles) B B is vrible condition nd A unifies with B P α () For { A Ak / B E β every condition in the set of { A Ak { B must follow the sme rule s () And then we denote how to implement the ction mtchmking in the following lgorithm Algorithm ActionMtchmking(); input: request ( F P ) nd P ( Pr Ef ) the set of cndidte services T = { n output: set of services whose ction is mtchmking with request function CheckActionubsumption(α β ) input: two ctions α nd β output: if α β return true else return flse on-functionl ttributes mtchmking In sense non-functionl ttributes semntics of semntic web service nd o semntics re synonyms The ttributes tht ffect o of web service re clssified into two types: common ttributes including executing rte price nd network bndwidth pplictionspecific ttributes tht depend on specific ppliction domins o semntics could be described by DL formul especilly the GRF of the concepts In the following definition we define the non-functionl ttributes mtchmking Definition 0: (the non-functionl ttributes mtchmking) In the semntic mtching of web services let ( F P ) be Web service nd = A = { Ai Aj i j n be the GRF of ( F P ) be request nd = B Bm = { Bi B j i j m be the GRF of ε be the concept miss error () hs Exct non-functionl ttributes mtching reltion with iff: ( A B M( B ( = ) () hs trong non-functionl ttributes mtching reltion with iff: ( A B M( B ( ) () hs Wek non-functionl ttributes mtching reltion with iff: (( A B M( B { ( A B M ( BA ))) ( A A B M ( BA ) ε) W (4) If it is out of ll the bove situtions hs ot- Mtch non-functionl ttributes mtching reltion with The following lgorithm denotes the non-functionl ttributes mtchmking Algorithm onfuncmtchmking( ε );

5 input: request ( F P ) nd = B B m concept miss error ε the set of cndidte services T = { n output: set of services whose nonfunctionl ttributes re mtchmking with request 4 emntic bsed Web service mtchmking Bsed on the three lgorithms bove the optiml service should be the service in the intersection of the service sets clculted by these lgorithms respectively If the intersection is null the service tht owns highest mtching degree is selected s the optiml service As mentioned in the mtchmking principle before in most cses the optiml service should stisfy the following three priority-rnked mtchmkings: functionl cpbilities mtchmking; ction mtchmking; nonfunctionl ttributes mtchmking Thus the mtching lgorithm of semntic web service is presented s follows: () functionl cpbilities mtching between the web service of the WL- librry nd the request is firstly performed the mtching result --- service set nmed T contins ll the services with the highest mtching rnk (in other words if the highest functionl cpbilities mtching rnk in the WL- librry is HR then the mtching rnk of ech service belonged to T equls to HR); (b) ction mtching or non-functionl mtching is performed ccording to the specific requirements in the request; (c) select the optiml service relied on the rules listed below: () prior selecting the service with the higher ction mtching degree if different ction mtching rnks exist () prior selecting the service with the higher nonfunctionl ttributes mtching rnk if different nonfunctionl ttributes mtching rnk exist on the premise of the sme ction mtching rnk () if two or more services hve the sme mtching rnk in ll three spects prior selecting the service with higher reputtion vlue (service reputtion is tken into considertion in this step becuse the finl discovery result closely reltes to it which would be described in detil in [8]) Algorithm ervicemtchmking( ε Domin); input: request ( F P ) concept miss error ε the correltive Domin of request the set of cndidte services T = { n output: set of services which re the most pproprite mtchmking with request begin service = ; for every i in T i n FuncMtchdgree = otmtch ; ActionMtchdgree = otmtch ; i i i onfuncmtchdgree = otmtch ; T = FunctionMtchmking( ε Domin); service = ChooseBestervices( T FuncMtch); if P is described then T = service ; T = ActionMtchmking(); service = ChooseBestervices( T ActionMtch); if then T = service ; T = onfuncmtchmking( ε ); service = ChooseBestervices( T onfuncmtch); if service = then return service; T = service ; for every i in T i Reputtion = ComputeReputtion( i ); service = ChooseBestervices(T Reputtion); return service; end function ChooseBestervices(T ctype) input: the set of services T the criterion ctype for selection output: pick out the services which hve the highest mtching rnk of ctype from T begin Resultet = ortbyctype(t); //sort the services in sorting order tht strts with the highest vlue of ctype nd proceeds to the lowest = the first element of T; for every i in T i n if the ctype vlue of i nd re equivlent then Resultet = Resultet { i ; return Resultet; end If the result of bove lgorithm is set contining multiple services with the sme mtching rnk nd reputtion the consumer my select service t rndom Besides to utomticlly invoke the web service the emntic Web service frmework my select rndom service s the trget which would chieve the sme effect s the requester selects service rndomly 40 Illustrting exmple In this section with n exmple we illustrte our pproch to semntic mtching of Web service There re request nd four cndidte services Figure depicts

6 the hierrchy of the domin ontologies bout the bicycle-selling service Let ( F P ) F ( I ) I VIACrd DeliverDte ( Titnium teel) MountinType ( Humn Electromotor) 00To00 Bckset ( hopdistnce0 km) ( BlnceInCC00) = ( BlnceInCC BlnceInCC theprice) P is not specified hsdeliverperioddys HomeDelivery () ( F P ) F ( I ) I MATERCrd VIPMn MountinType Electromotor More00 Bbyset ( BlnceInCC500) = ( BlnceInCC BlnceInCC theprice) HomeDelivery () ( F P ) F ( I ) I CreditCrd Titnium MountinType ( Electromotor Humn ) 00To 00 ( hopdistnce0 km) = ( BlnceInCC BlnceInCC theprice) hsdeliverperioddys () ( F P ) F ( I ) I CreditCrd DeliverDte 8ize Titnium MountinType Humn 00To00 Bckset Bbyset ( hopdistnce0 km) ( BlnceInCC00) = ( BlnceInCC BlnceInCC theprice) hsdeliverperioddys HomeDelivery (4) 4 ( F P ) 4 4 F 4 ( I ) I CreditCrd DeliverDte 4 4 6ize teel MountinType Electromotor 00To00 Bckset ( hopdistnce0 km) ( BlnceInCC00) 4 = ( BlnceInCC BlnceInCC theprice 4 HomeDelivery 4 ) Let concept miss error ε = In there re concepts in I not be mtched by I nd concepts in not be mtched by so does not mtch in functionl cpbilities There is just one concept in not be mtched by so hs wek functionl cpbilities mtching reltion with For the rest nd 4 ll hve strong mtching rnk with in the functionl cpbilities mtchmking Also becuse P is not specified we should next perform non-functionl ttributes mtchmking between 4 nd It is esy to see tht hs exct mtching rnk wheres 4 only hs strong mtching rnk in non-functionl ttributes mtchmking Finlly the optiml service which hs highest mtching rnk with is tyleize 6ize 8ize 6ize Humn Electromotor Humn 8ize Electromotor Mteril teel Titnium teel FuncType MountinType FltType MountinType Motivity Humn Electromotor Petrolic Titnium Price Less00 00To00 More00 Less00 00To00 More00 Petrolic Electromotor Petrolic VIACrd CreditCrd FltType Bckset Bbyset Bell CreditCrd Pyment DeliverDte hsdeliverperioddys HomeDelivery MATERCrd Fig The hierrchy of the domin ontologies bout the bicycle-selling service 50 Relted work Current Web services infrstructure for exmple UDDI provides limited serch fcilities llowing only keyword-bsed serch of services To cope with this limittion emerging pproches rely on emntic Web technology to support service discovery [9] For exmple Chkrborty et l [0] define n ontology bsed on DAML [] to describe mobile devices nd propose mtching mechnism tht loctes devices bsed on their fetures (eg type of printer) [] proposes to use process ontologies to describe the behvior of services nd then to query such ontologies using process query lnguge (PL) There re other pproches bsed on DAML+IL [] description of services tht propose to exploit the DL-bsed resoning mechnisms An experience in building mtchmking prototype bsed on DL resoner tht considers DAML+IL-bsed service descriptions is reported by [4] The proposed mtchmking lgorithm is bsed on simple subsumption nd consistency tests A more sophisticted mtchmking lgorithm between services nd requests described in DAML- is proposed by [9] The lgorithm considers vrious degrees of mtching tht re determined by the miniml distnce between concepts in the concept txonomy [5] presents n pproch to tckling this problem in the context of description logics They formlize service discovery s new instnce of the problem of rewriting concepts using terminologies which is clled the best covering problem Also novel mtchmking lgorithm is provide to find set of services clled best cover of request whose descriptions contin s much common informtion with request s possible nd s little extr informtion with respect to request s possible

7 60 Conclusion We hve ddressed the problem of mtchmking of web services from knowledge representtion perspective We hve presented suitble definitions of the problem The semntic informtion of Web service cn be described with three fctors: functionl cpbilities ction nd non-functionl ttributes We formlize these three fctors mtchmking nd propose novel mtchmking lgorithm We proposed mtch ctegoriztion in terms of Exct mtch trong mtch Wek mtch nd ot- Mtch nd rnk of mtches within ctegories At lst we hve presented simple exmple to illustrte our pproch ur future work will be devoted to perform the more precise semntic mtching of Web service ie quntify the mtchmking degree in order to improve precision nd recll of the service discovery [] HorrocksI (00) DAML+IL: resonble Web ontology lnguge In: Proceedings of EDBT 00 Prgue Czech Republic Mrch 00 pp - [4] Gonz lez-cstillo J Trstour D Brtolini C (00) Description logics for mtchmking of services In: Proceedings of the KI-00 workshop on pplictions of description logics Vienn Austri eptember 00 de/publictions/ceur-w/vol-44/ [5] B Bentllh et l (004) n utomting Web services discovery VLDB Journl (004) / Digitl bject Identifier (DI) 0007/s x 70 References [] DAML ervices Colition (00) DAML-: Web service description for the emntic Web In: Proceedings of the st interntionl emntic Web conference (IWC) rdini Itly June 00 pp 48 6 [] Fensel D Bussler C MedcheA (00) emntic Web enbled Web services In: Proceedings of the interntionl emntic Web conference rdini Itly June 00 pp - [] Donini F cherf A Lenzerini M rdi D (996) Resoning in description logics In: Brewk G (ed) Foundtion of knowledge representtion CLI- Publictions tnford CA pp 9-6 [4] HorrocksI Ptel-chneider PF vn Hrmelen F (00) Reviewing the design of DAML+IL: n ontology lnguge for the emntic Web In: Proceedings of the 8th ntionl conference on rtificil intelligence (AAAI 00) Edmonton Albert Cnd 8 July August 00 pp [5] World Wide Web Consortium (00) [6] World Wide Web Consortium (004) [7] Zhongzhi hi Mingki Dong et l (004) The logic bse of emntic Web service cience in chin series E: Informtion ciences 004 Vol4 o0 pp-8 [8] Dingjin Chen Jin Wu et l (005) Design nd implementtion of semntic Web service reputtion model Journl of computer pplictions Chin 005 Vol5 o8 pp [9] Polucci M Kwmur T Pyne TR ycr KP (00) emntic mtching of Web services cpbilities In: Proceedings of the interntionl emntic Web conference rdini Itly June 00 pp 47 [0] Chkrborty D Perich F Avnch Joshi A (00) DReggie: semntic service discovery for M-Commerce pplictions In: Proceedings of the workshop on relible nd secure pplictions in mobile environment 0th symposium on relible distributed systems ew rlens ctober 00 pp 8- [] DAML ervices [] Bernstein A Klein M (00) Discovering services: towrds high precision service retrievl In: Proceedings of the CiE workshop on Web ervices e-business nd the emntic Web: foundtions models rchitecture engineering nd pplictions Toronto My 00