On User Preferences and Utility Functions in Selection: A Semantic Approach

On Use P e e ences and U ili y Func ions in Selec ion: A Seman ic App oach Jos´eMa ´ıa Ga c´ıa, Da id Ruiz, and An onio Ruiz-Co ´es Uni e sidad de Se illa Escuela T´ecnica Supe io de Ingenie ´ıa In o m´a ica A . Reina Me cedes s/n, 41012 Se illa, Espa˜na [email p o ec ed] Abs ac . Disco e y asks in he con ex o Seman ic Web Se ices a e gene ally pe o med using Desc ip ion Logics. Howe e , his o malism is no sui ed when non- unc ional, nume ical pa ame e s a e in ol ed in he disco e y p ocess. Fu he mo e, in selec ion asks, whe e an op- imiza ion algo i hm is needed, DLs a e no capable o compu ing he op imum. Al hough he e a e DLs ex ensions ha can handle nume ical pa ame e s, hey b ing decidabili y p oblems. O he solu ions, as hyb id app oaches which use DLs in unc ional disco e y and o he o malisms in non- unc ional selec ion, do no p o ide a seman ic amewo k o de- sc ibe use p e e ences based on non- unc ional p ope ies. In his wo k, we p opose o seman ically desc ibe use p e e ences, so hey can be used o pe o m selec ion wi hin a hyb id solu ion. By using seman ically de- sc ibed u ili y unc ions in o de o define use p e e ences, ou p oposal enables in e ope abili y be ween se ice offe s and demands, while p o- iding a high le el o exp essi eness in hese p e e ences and including hem wi hin SWS desc ip ions. Keywo ds: NFP-based Selec ion, Quali y o Se ices, U ili y Func ions, Seman ic Web Se ices. 1 In oduc ion Conce ning Seman ic Web Se ices (SWS), disco e y is one o he main esea ch opics ha ha e been widely s udied and discussed, among o he s like compo- si ion. Desc ip ion Logics (DLs) usually ha e become he na u al choice when disco e ing SWS. T adi ionally, disco e y asks ha e been in e p e ed as a unc- ional fil e , whe e demands a e ma ched wi h compa ible offe s in e ms o unc- ionali y. Howe e , including non- unc ional p ope ies (NFP) in he disco e y p ocess leads o an op imiza ion p oblem. Selec ion o he bes offe by means o hei NFP has no been con empla ed as a main ask in disco e ing, so DLs easone s a e no well sui ed o selec op imal offe s. Howe e , he e a e some p oposals o pe o m NFP-based disco e y, as he ones discussed in his wo k. This wo k has been pa ially suppo ed by he Eu opean Commission (FEDER) and Spanish Go e nmen unde CICYT p ojec Web-Fac o ies (TIN2006-00472). Op imiza ion p oblems can be handled by sol e s based on diffe en o - malisms, like Linea P og amming, Cons ain P og amming, o Dynamic P o- g amming, among o he s. Thus, i is possible o spli disco e y and selec ion asks in e ms o unc ional and non- unc ional equi emen s, so he o me ask can be pe o med by DLs easone s, while he la e can be pe o med by sol e s, aking a hyb id app oach [3]. Focusing on selec ion, se ice demands ha e o s a e an op imali y c i e ion, i.e. use p e e ences, so a sol e can ob ain he bes offe in e ms o hese p e e ences. We p opose o desc ibe hese use p e e ences by means o u ili y unc ions, whose domains a e he diffe en QoS pa ame e s used o define NFP o se ice offe s. In a SWS con ex , hese u ili y unc ions ha e o be seman ically desc ibed, allowing o ma ch demands and offe s desc ibed by diffe en , bu possibly equi alen , QoS pa ame e s, enabling seman ic in e ope abili y be ween hese desc ip ions. The pape is s uc u ed as ollows. In Sec. 2 we analyze cu en app oaches on selec ing SWS. Then, in Sec. 3 we show ou p oposal, desc ibing wha is a u ili y unc ion and how o gi e seman ics o i , showing an example. Finally, in Sec. 4 we discuss ou conclusions. 2 Selec ing Seman ic Web Se ices Once a se o se ices a e disco e ed using a unc ional fil e , he nex s ep is o selec he bes offe in e ms o NFP and use p e e ences. Thus, selec ion is modeled as an op imiza ion p oblem. This kind o p oblem e e s o a minimiza- ion o maximiza ion o a eal unc ion, choosing he app op ia e alues o he in ol ed a iables. In his con ex , he unc ion o op imize is equen ly called u ili y unc ion o objec i e unc ion. The e a e diffe en echniques o ob ain he op imal alue o hese unc ions, like Linea P og amming, Cons ain P og amming o Dynamic P og amming, o ins ance. In he ollowing, we p esen he diffe en app oaches on selec ing offe s by means o NFP, cha ac e izing hei ea u es and limi a ions wi h espec o use p e e ences. 2.1 Cu en App oaches An ea ly app oach on modeling QoS in he con ex o SWS disco e y a e ound in[10].In hiswo k,Ranp esen sauddi ex ension and a ca alog o QoS pa am- e e s ha can be included in uddi desc ip ions. Disco e y is pe o med using que ies wi h unc ional equi emen s,aswellascondi ionsonQoS. Howe e , he ac ual disco e y algo i hm is no defined, and que ies ha use NFP a e no shown, so hei exp essi eness is unknown. Addi ionally, uddi only suppo s a keywo d based sea ch, so no o m o in e ence o flexible ma ch can be pe o med [15]. Apa om ha , use p e e ences can no be exp essed in he que y and he esul an se ices a e no anked, so he use ha e o pe o m diffe en que ies in o de o find he bes sui ed se ice. Al hough hei p oposal is no seman ically defined, Liu e al. p esen a QoS compu a ion model including a selec ion algo i hm [5], which is adap ed in o he app oaches [9,16]. They p opose an ex ensible QoS model ha comp ises bo h gene ic and domain specific c i e ia. Selec ion is pe o med using an algo i hm based on ma ices no maliza ion, whe e se ices a e anked in e ms o hei QoS ma ix desc ip ion and a ec o o ela i e weigh s be ween QoS pa ame e s, which exp ess use p e e ences. Pa hak e al. also model mappings be ween on ologies in [9]. They p opose o use domain specific on ologies o define NFP among offe s and demands. In hei wo k, selec ion is done using ma ching deg ees a a fi s s age. Then, QoS pa ame e s alues a e collec ed in a quali y ma ix, which is used o calcula e a fixed, weigh ed u ili y unc ion o each offe . Finally, offe s whose u ili y unc ion is abo e a gi en h eshold, a e anked by one QoS pa ame e o ob ain he op imal offe . Wang e al. p o ide an ex ension o wsmo on ology [11] o handle QoS pa- ame e s [16]. They define a QoS selec ion model and an algo i hm based on a quali y ma ix ha con ains alues o QoS pa ame e s. The use p e e ences a e desc ibed in e ms o endencies, i.e. a demand may p e e pa ame e s o be as small as possible, as la ge as possible, o a ound a gi en alue. Maximilien and Singh p esen a amewo k and a QoS on ology o dynamic selec ion in [8]. They use an agen -based app oach whe e NFP a e modeled ia a h ee-laye on ology: an uppe on ology which defines basic concep s associa ed wi h a quali y pa ame e , a middle on ology which defines he mos equen QoS pa ame e s and me ics, and a use -defined lowe on ology ha depends on he domain o he se ice. Al hough i cons i u es a well-defined amewo k o seman ically desc ibe NFP and i is e e enced by many au ho s [2,4,9], i lacks o a way o seman ically desc ibe use p e e ences. An ex ension o daml-s 1 o include QoS p ofiles is p oposed in [18] by Zhou e al. This p oposal only allows o de condi ions be ween QoS pa ame e s, so i pe o ms disco e y and selec ion using DLs. The QoS on ology is simple and can be easily linked o he daml-s se ice p ofile. Howe e , i s selec ion algo i hm uses ma ching deg ees o o de he esul ing se o se ices, so he use p e e ences can no be exp essed, as hey a e inhe en o ha selec ion algo i hm. Ano he daml-based p oposal is also p esen ed in [14], whe e S. Bilgin and Singh p o ide a daml-based que y language, ins ead o jus ex ending owl- s. Using his Seman ic Web Se ices Que y and Manipula ion Language, hey ad e ise QoS a ibu es and pe o m he selec ion. The main d awbacks o his app oach a e he same as in [10], wi h limi a ions on he exp essi eness o que ies, due o he use o daml as i s ounda ion. Thus, use p e e ences can no be exp essed in hose que ies, and a e inhe en o hei selec ion algo i hm, as in [18]. Dobson e al. p esen s QoSOn in [2], which is an on ology ha ex ends owl-s o desc ibe QoS a ibu es and me ics. Howe e , hey do no explici ly explain how o pe o m selec ion, and hei p oposal suffe s om owl limi a ions, so 1daml-s is an ea ly e sion o owl-s [6]. hey ha e o use an ad-hoc XML language o allow cus om da a anges. Use p e e ences a e modeled using he accep abili y di ec ion, ha is he p e e ed endency o me ic alues (e.g. he highe he bes ). On he o he hand, Zeng e al. show a basic QoS model o Web se ices composi ion in [17], al hough i can be applied o disco e y and selec ion. They p opose an algo i hm based on u ili y unc ions, which a e al eady defined o all he con empla ed QoS pa ame e s. The op imiza ion is implemen ed using In ege P og amming, p o iding weigh s o he diffe en QoS pa ame e s in- ol ed. The main d awbacks o his p oposal a e ha i do no ake seman ics in o accoun and ha he u ili y unc ions a e fixed, so he use can define i s p e e ences only by means o weigh s. Ruiz-Co ´es e al. desc ibe a QoS-awa e disco e y using Cons ain P og am- ming, whe e op imiza ion is modeled as a Cons ain Sa is ac ion Op imiza ion P oblem ha minimize a weigh ed composi ion o u ili y unc ions, which a e defined by he clien using QoS pa ame e s om a ca alog [12]. As in [17], his p oposal does no p o ide seman ics, bu use p e e ences, desc ibed by u ili y unc ions, can be defined by he use wi h high exp essi eness. An ex ension o [7] is p esen ed in [4] by K i ikos and Plexousakis. They p opose an on ology simila o he p oposed by Maximilien and Singh [8], mix- ing offe s and demands wi hin an owl-s desc ip ion. Mo eo e , hey p esen a ma ching algo i hm o in e equi alences be ween diffe en named QoS pa- ame e s ha a e seman ically equi alen , al hough i is gene ally undecidable. Conce ning disco e y and selec ion, hey use CSPs o pe o m he ma chmaking o compa ible offe s, and hen selec he bes se ice by means o a weigh ed composi ion o u ili y unc ions, which balance he wo s and bes scena ios o compu e he u ili y alue. Howe e , hese use p e e ences a e no seman ically defined in hei QoS on ology. 2.2 Analysis We show an analysis o he ea u es o he diffe en app oaches in oduced in he sec ion be o e in Table 1. In his able, o de ed by he o de o exposi ion, we analyze i he gi en p oposal seman ically defines NFP, how i exp ess use p e e ences, and he selec ion algo i hm used. We ob ain se e al conclusions om his compa ison. Fi s ly, he e a e a ew p oposals ha uses u ili y unc ions o exp ess use p e e ences [4,12,17], al- hough only [12] allows he use o define complex u ili y unc ions. These h ee p oposals use op imiza ion echniques, as In ege P og amming o Cons ain P og amming, o selec he bes offe s. The e o e, u ili y unc ions become he na u al choice o define highly exp essi e use p e e ences. Secondly, he e a e many p oposals ha p o ide a seman ic amewo k o define NFP [2,4,8,9,14,16,18], al hough [8] do no handle use p e e ences in hei on ology and [14,18] ha e a fixed defini ion o use p e e ences, inhe en o hei selec ion algo i hm. [4] is he mos exp essi e when defining use p e e - ences, ollowed by [2,9,16], ha limi hei p e e ences o weigh s and pa ame e Table 1. Compa ison be ween discussed p oposals P oposal Seman ic De s. Use P e e ences Selec ion Ran [10] No No defined No defined Liu e al. [5] No Weigh s Quali y ma ix Pa hak e al. [9] Yes Weigh s Quali y ma ix Wang e al. [16] Yes Tendencies Quali y ma ix Maximilien & Singh [8] Yes Ex e nal Ma ching deg ee Zhou e al. [18] Yes Fixed Ma ching deg ee S. Bilgin & Singh [14] Yes Fixed Que y lang. Dobson e al. [2] Yes Tendencies No defined Zeng e al. [17] No U ili y and weigh s In ege P og. Ruiz-Co ´es e al. [12] No U ili y and weigh s Cons ain P og. K i ikos & Plexousakis [4] Yes U ili y and weigh s Cons ain P og. endencies. Acco ding o all hose p oposals, i is clea ha NFP ha e o be defined seman ically. Finally, we conclude ha none o he abo e p oposals seman ically define use p e e ences, al hough in [2,16] he au ho s include in hei on ology ex ension he endency o QoS pa ame e s. Wha is mo e, mos o he p oposals ha pe o m selec ion asks in e ms o use p e e ences desc ibe hem using ad-hoc, non-seman ic desc ip ions comple ely decoupled wi h he ones used o desc ibe se ice unc ionali y, causing a seman ic gap be ween unc ional desc ip ions and use p e e ences. The mo i a ion o ou wo k is p ecisely o ackle he p e ious p oblems. Mos ecen p oposals use u ili y unc ions o exp ess use p e e ences, and he e a e many NFP on ologies which ou p oposal can be in eg a ed wi h. This p oposal comes om mixing he exp essi eness o u ili y unc ions and weigh s p oposed by Ruiz-Co ´es e al., he seman ic defini ion o NFP om Maximilien and Singh o K i ikos and Plexousakis, and an ex ension o gi e seman ics o u ili y unc- ions. Thus, we ake ull ad an age o Seman ic Web app oaches on selec ing he bes se ices, while allowing o define use p e e ences using he mos exp essi e solu ion, i.e. u ili y unc ions. Fu he mo e, we pu unc ional, non- unc ional, and use p e e ences a he same seman ic le el, by means o using ex ensions o cu en SWS on ologies. 3 Ou P oposal U ili y unc ions a e he mos exp essi e app oach p esen ed o desc ibe use p e e ences ha a e used when selec ing he bes offe s among a se . Al hough he e a e p oposals ha seman ically desc ibe QoS pa ame e s and NFP, no one con empla es he concep ualiza ion o u ili y unc ions. In his Sec ion, we fi s ly gi e an o e iew o u ili y unc ions. Then, an on ology o use p e e ences is p oposed o be used in he con ex o disco e y and selec ion o SWS, showing a conc e e example. 3.1 U ili y Func ions An u ili y unc ion is a no malized unc ion ( anging o e [0,1]) whose domain is a gi en QoS pa ame e , ha gi es in o ma ion abou which alues o ha QoS pa ame e a e p e e ed by he use . Fig. 1 shows an example o a u ili y unc ion o he mean ime o ailu e (MTTF) pa ame e . This unc ion is a piecewise linea unc ion ha defines a minimum u ili y (0) when MTTF is below 60 minu es, and a maximum u ili y (1) when MTTF is abo e 120 minu es. Be ween hese wo limi s, he unc ion g ows linea ly. Fig. 1. U ili y example o Mean Time To Failu e When selec ing he bes offe s, a composi ion o diffe en u ili y unc ions (one o each QoS pa ame e in ol ed in NFPs) is o en used o compu e a global u ili y alue, which se es o so he se ice offe s. In his composi ion, each u ili y unc ion has an associa ed weigh in he global unc ion, so he use can speci y how impo an is each QoS pa ame e when selec ing offe s. The gene al o m o his weigh ed composi ion o u ili y unc ions Uis as ollows [12]: U(p1,...,p n)= n  i=1 kiUi(pi),k i∈[0,1] n  i=1 ki=1 (1) whe e each pideno es a QoS pa ame e , each kii s associa ed weigh anging o e [0,1], and each Uii s associa ed u ili y unc ion also anging o e [0,1] wi h he seman ics p e iously defined. 3.2 Gi ing Seman ics o U ili y Func ions In o de o p o ide seman ic in e ope abili y be ween u ili y unc ions defined on diffe en ly named QoS pa ame e s, we p opose o model hese unc ions, o mo e gene ally, use p e e ences, as an on ology. This on ology has o be ins an ia ed by each use p e e ence desc ibing u ili y unc ions, so equi alences be ween QoS pa ame e s can be in e ed. Fu he mo e, his concep ualiza ion allows he use o desc ibe he whole se ice, including unc ional desc ip ions, a he same seman ic le el, wi hou coupling use p e e ences desc ip ions wi h he selec ion algo i hm. Ou p oposed model is shown in Fig. 2. The main concep (o class)isUse - P e e ence, which e e ences a Quali y concep ia he hasRe e ence objec p op- e y. This Quali y concep is analogous o he defined in [8], and ep esen s he QoS pa ame e which is used in he defini ion o he co esponding use p e - e ence. Fu he mo e, he Use P e e ence concep has a key da a ype p ope y, hasDefini ion, which links he mo e gene ic p e e ence concep wi h he u ili y unc ion ha defines i . No e ha Quali y class is he link o QoS pa ame e s used in he seman ic defini ion o NFP. This defini ion can be pe o med using he on ology om K i ikos and Plexousakis [4] o om Maximilien and Singh [8], o ins ance. Class Use P e e ence Class Quali y ange domain ange domain Da aTypeP ope y hasDe ini ion ange domain S ing Floa domain ange Objec P ope y hasRe e ence Da aTypeP ope y hasName Da aTypeP ope y hasWeigh XMLLi e al Fig. 2. P oposed on ology o model use p e e ences The u ili y unc ion is ini ially modeled as a p ope y ha con ains an XML exp ession ha desc ibe he defini ion o each unc ion in e ms o OpenMa h s anda d [1], as used in [4,13], allowing he e alua ion o he unc ion wi h a p ope compile o a ma hema ical ool, such as Ma hema ica. Finally, ou main concep Use P e e ence has wo da a ype p ope ies: has- Name and hasWeigh . The o me is used as an iden ifie o a gi en ins ance. The la e is a eal numbe which co esponds o he ela i e weigh associ- a ed wi h he co esponding QoS pa ame e , used o compu e he global u ili y unc ion o an offe . Fig. 3 shows an ins ance o ou p oposed on ology, in he case o an use p e e ence abou MTTF, wi h an associa ed weigh o 1. Thus, he ins ance MTTFUse P e e ence e e ences an ins ance Use MTTF o MTTF class, ha is asubclasso Quali y class om [8]. Mo eo e , he conc e e u ili y unc ion is specified as an OpenMa h objec ha ep esen he one showed in Fig. 1, using XML. hasRe e ence hasName “MTTF_UP” 1.0 hasWeigh hasDe ini ion MTTFUse P e e ence Use MTTF Class Use P e e ence Class MTTF ins anceO ins anceO <om:OMOBJ> </om:OMOBJ Class Quali y subClassO Fig. 3. Ins ance o ou p oposed on ology The link wi h he es o he seman ic desc ip ion o a se ice, including bo h unc ional and non- unc ional p ope ies, is he QoS pa ame e MTTF, i.e Use MTTF ins ance in ou example. In his case, he engine ha pe o m he selec ion only has o be awa e o he pa showed in Fig. 3 and he co - esponding NFPs ha in ol e he MTTF pa ame e , bu gene ally, he e a e mo e pa ame e s and use p e e ences in ol ed in he selec ion p ocess. 4 Conclusions In his wo k, we p o ide a seman ic amewo k o define use p e e ences on seman ically defined QoS pa ame e s, p o ided ha i is used in conjunc ion wi h ano he p oposal ha seman ically defines NFP, like [4,8]. Thus, all ace s o SWS desc ip ion ( unc ional, non- unc ional, and use p e e ences) a e desc ibed a he same seman ic le el, so disco e y and selec ion asks a e comple ely done wi hin a single seman ic amewo k, allowing in e ope abili y be ween diffe en se ice defini ions. Fu he mo e, ou p oposal uses a e y exp essi e solu ion o define use p e - e ences, i.e. u ili y unc ions and weigh s, as in [4,12,17]. This o malism becomes mo e gene ic and powe ul ha ones used in o he app oaches. Addi ionally, we p opose he use o a hyb id disco e y engine o pe o m he diffe en asks in disco e y and selec ion using he bes sui ed echnique in each case. Thus, we op imize hese asks wi hou comp omising exp essi eness. In conclusion, ou p oposal allows a seman ic defini ion o he whole disco e y and selec ion p ocess, using a hyb id app oach wi hou losing exp essi eness. These ac s allow o decouple he defini ion o use p e e ences om he conc e e selec ion algo i hm used. Acknowledgmen s. The au ho s would like o hank he e iewe s o he NFPSLA-SOC’07 Wo kshop, whose commen s and sugges ions imp o ed he p esen a ion subs an ially. Re e ences 1. Buswell, S., Cap o i, O., Ca lisle, D.P., Dewa , M.C., Ga¨ e ano, M., Kohlhase, M.: The OpenMa h s anda d. Technical Repo Ve sion 2.0, The OpenMa h Socie y (2004) 2. Dobson, G., Lock, R., Somme ille, I.: Qoson : a qos on ology o se ice-cen ic sys ems. In: EUROMICRO-SEAA, pp. 80–87. IEEE Compu e Socie y P ess, Los Alami os (2005) 3. Ga c´ıa, J.M., Ruiz, D., Ruiz-Co ´es, A., Ma ´ın-D´ıaz, O., Resinas, M.: An hyb id, QoS-awa e disco e y o seman ic web se ices using cons ain p og amming. In: K ¨ ame , B.J., Lin, K.-J., Na asimhan, P. (eds.) ICSOC 2007. LNCS, ol. 4749, pp. 69–80. Sp inge , Heidelbe g (2007) 4. K i ikos, K., Plexousakis, D.: Seman ic QoS me ic ma ching. In: ECOWS 2006, pp. 265–274. IEEE Compu e Socie y P ess, Los Alami os (2006) 5. Liu, Y., Ngu, A.H.H., Zeng, L.: Qos compu a ion and policing in dynamic web se ice selec ion. In: WWW (Al e na e T ack Pape s & Pos e s), pp. 66–73 (2004) 6. Ma in, D., Bu s ein, M., Hobbs, J., Lassila, O., Mcde mo , D., e al.: OWL-S: Seman ic Ma kup o Web Se ices. Technical Repo 1.1, DAML (2004) 7. Ma ´ın-D´ıaz, O., Ruiz-Co ´es, A., Bena ides, D., Du ´an, A., To o, M.: A quali y- awa e app oach o web se ices p ocu emen . In: 4 h VLDB Wo kshop on Tech- nologies o E-se ices TES 2003, pp. 42–53 (2003) 8. Maximilien, E.M., Singh, M.P.: A amewo k and on ology o dynamic web se ices selec ion. In e ne Compu ing 8(5), 84–93 (2004) 9. Pa hak, J., Koul, N., Ca agea, D., Hona a , V.G.: A amewo k o seman ic web se ices disco e y. In: WIDM 2005: P oceedings o he 7 h annual ACM in e na- ional wo kshop on Web in o ma ion and da a managemen , New Yo k, NY, USA, pp. 45–50. ACM P ess, New Yo k (2005) 10. Ran, S.: A model o web se ices disco e y wi h QoS. SIGecom Exch. 4(1), 1–10 (2003) 11. Roman, D., Lausen, H., Kelle , U.: Web Se ice Modeling On ology (WSMO). Technical Repo D2 1.3 Final D a , WSMO (2006) 12. Ruiz-Co ´es, A., Ma ´ın-D´ıaz, O., Du ´an-To o, A., To o, M.: Imp o ing he au o- ma ic p ocu emen o web se ices using cons ain p og amming. In . J. Coope - a i e In . Sys . 14(4), 439–468 (2005) 13. S´anchez-Maci´an, A., L´opez, D., L´opez de Ve ga a, J.E., Pas o , E.: A amewo k o he au oma ic calcula ion o quali y o expe ience in elema ic se ices. In: 13 h HP-OVUA Wo kshop, Sophia An ipolis, F ance (May 2006)