Ranking Optimization with Constraints

Size: px
Start display at page:

Download "Ranking Optimization with Constraints"

Transcription

1 Rakig Opimizaio wih Cosrais Fagzhao Wu, Ju Xu, Hag Li, Xi Jiag Tsighua Naioal Laboraory for Iformaio Sciece ad Techology, Deparme of Elecroic Egieerig, Tsighua Uiversiy, Beijig, Chia Noah s Ark Lab, Huawei Techologies Co. Ld., Sha Ti, Hog Kog wufagzhao@gmail.com, juxu@ic.ac.c, {hagli.hl, jiag.xi}@huawei.com ABSTRACT This paper addresses he problem of pos-processig of rakig i search, referred o as pos rakig. Alhough impora, o research seems o have bee coduced o he problem, paricularly wih a pricipled approach, ad i pracice ad-hoc ways of performig he ask are beig adoped. This paper formalizes he problem as cosraied opimizaio i which he cosrais represe he pos-processig rules ad he objecive fucio represes he rade-off bewee adherece o he origial rakig ad saisfacio of he rules. The opimizaio amous o refiig he origial rakig resul based o he rules. We furher propose a specific probabilisic implemeaio of he geeral formalizaio o he basis of he Bradley-Terry model, which is heoreically soud, effecive, ad efficie. Our experimeal resuls, usig bechmark daases ad eerprise search daase, show ha he proposed mehod works much beer ha several baselie mehods of uilizig rules. Caegories ad Subjec Descripors H.3.3 [Iformaio Sorage ad Rerieval]: Iformaio Search ad Rerieval Rerieval models Keywords Pos Rakig; Rakig Opimizaio; Bradley-Terry Model 1. INTRODUCTION Rece years have observed a sigifica progress i research ad developme o learig o rak, i.e., creaio of rakig models i search usig machie learig echiques. Now, i becomes a commo pracice o exploi he learig echologies o cosruc he basic rakig model of a search sysem. This paper is cocered wih pos processig of rakig, which we call pos rakig. Pos rakig Currely affiliaed wih Isiue of Compuig Techology, Chiese Academy of Scieces. Permissio o make digial or hard copies of all or par of his work for persoal or classroom use is graed wihou fee provided ha copies are o made or disribued for profi or commercial advaage ad ha copies bear his oice ad he full ciaio o he firs page. Copyrighs for compoes of his work owed by ohers ha ACM mus be hoored. Absracig wih credi is permied. To copy oherwise, or republish, o pos o servers or o redisribue o liss, requires prior specific permissio ad/or a fee. Reques permissios from permissios@acm.org. CIKM 14, November 03-07, 2014, Shaghai, Chia. Copyrigh 2014 ACM /14/11...$ hp://dx.doi.org/ / is ormally coduced a web search egies i ad-hoc maers. The paper aims o provide a pricipled approach o pos rakig, which does o seem o have bee seriously sudied previously. I pracice, here are may siuaios i which oe was o furher wis he search resuls give by he basic rakig model, i.e., o coduc pos rakig. Pos rakig is widely adoped i pracice, uder he ames of re-rakig, fial rakig, ec. For example, he query is abou a ho opic ad oe was o boos a webpage abou he opic from ews chaels o he op hree posiios, o maer how he rakig model does (oe ha i is usually hard o add such corol io a learig o rak model). I aoher example, a web page is repored o be likely a spam page, ad a immediae acio is required o demoe he posiio of he page, wihou chage of he rakig model. (See more examples i Secio 3.) Pos rakig eeds o be carried ou o oly from he viewpoi of ehacig search qualiy, bu also due o operaioal, commercial, ad eve poliical reasos. There are oher siuaios i which pos rakig appears o be ecessary, such as diversificaio of search resul [6, 27, 7], coex aware rakig [29], persoalized rakig [25, 26, 23]. Therefore, pos rakig is a ecessary ad impora process for search. Pos rakig has he followig characerisics. (1) I is usually query depede, user depede, or coex depede. (2) The effecs of i may o be achieved by usig he basic rakig model, because i is usually difficul, cosly, or eve impossible o impleme i he basic rakig model. (3) I does o eed learig or raiig. The key challege for pos rakig lies i he difficuly of formalizig he problem i a heoreically soud, effecive, ad efficie way. The origial search resul migh be very differe from he rules of pos processig, ad he rules migh also be coradicory o each oher. Thus, i is o easy o icorporae he complicaed corols io a sigle framework. Moreover, he process eeds o be coduced olie ad hus mus be very efficie. This paper proposes formalizig rakig opimizaio as a cosraied opimizaio problem. Give he rakig resul of a query by he rakig model, we perform re-rakig o he resul, by miimizig a objecive fucio uder a umber of cosrais, where he cosrais represe he rules which we wa o use for pos processig, for example, boos a page o he op k posiio, ad he objecive fucio represes he rade-off bewee adherece o he origial rakig ad saisfacio of he cosrais.

2 As he firs sudy, we propose a mehod for rakig opimizaio. Our mehod adops he Bradley-Terry model [2] for calculaig he probabiliy of a rakig lis. I realizes he opimizaio problem as miimizig he egaive log codiioal probabiliy of he origial rakig lis ad he egaive log codiioal probabiliy of he cosrais give a Bradley-Terry model. The opimizaio problem has a simple form ad is guaraeed o have a global opimal soluio. Our mehod employs gradie dece o fid he opimal soluio, wih liear order ime complexiy. I is hus a heoreically soud, effecive, ad efficie mehod. We have coduced experimes usig he LETOR bechmark daases [14, 22] ad a daase from a eerprise search egie. We ake as baselies several mehods which adjus rakig resuls wih heurisic rules. Experimeal resuls idicae ha our mehod cosisely ad sigificaly ouperform he baselie mehods i erms of NDCG o all daases, idicaig ha i is beer o employ our mehod i pos rakig. The coribuios of he paper iclude (1) formalizaio of rakig opimizaio, (2) proposal of a mehod of rakig opimizaio, (3) empirical sudy of rakig opimizaio. The res of he paper is orgaized as follows. Afer a iroducio o relaed work i Secio 2, we describe he formulaio of rakig opimizaio wih cosrais i Secio 3. We describe he proposed mehod based o Bradley-Terry model i Secio 4. Experimeal resuls ad discussios are give i Secio 5. Secio 6 cocludes his paper ad gives fuure work. 2. RELATED WORK Cosrucio of rakig model is oe of he key problems i IR. Give a query, he rakig model assigs relevace scores o he rerieved documes ad sors he documes based o he scores, ad hus i plays a impora role i search. Tradiioally, a rakig model is defied based o a small umber of facors, e.g., erm frequecy, iversed docume frequecy, ad docume legh. BM25 [24] ad LMIR (Laguage Models for Iformaio Rerieval) [21, 12] are such models. Recely, machie learig echiques are applied o cosrucio of rakig model usually usig a large umber of feaures ad a large amou labeled raiig daa, referred o as learig o rak. Mehods of learig o rak are caegorized as poiwise [18, 13], pairwise [9, 8, 3, 4], ad liswise [5, 30, 31, 32, 28] mehods, depedig o he loss fucios used. I his paper, we adop he liswise mehod of [28] o rai he basic rakig model. As explaied above, basic rakig is o eough, ad pos rakig is ecessary i may cases. I pracice, pos rakig is coduced by usig heurisic rules, ad here has o bee research o pos rakig iself, as far as we kow. There are several oher rakig issues, which ca be addressed hrough pos rakig as well, such as search resul diversificaio, persoalized rakig, ad coex aware rakig. I rece years, search resul diversificaio arises as a ho opic i IR, i which he search sysem reurs a lis of documes which are o oly releva o he query bu also cover may subopics of he query. A commo pracice for diversificaio is o see diversificaio as a pos rakig sep afer he rakig lis based o relevace is creaed [6, 27, 7]. This is because i is usually hard o model relevace ad diversiy i a sigle framework. Persoalized search may also be realized i pos rakig, for example, i wha is called clie side re-rakig [25, 26, 23]. Specifically, he rakig resul by he basic rakig model is se o he clie side ad re-rakig of he resul based o he user s ieres is coduced. Oe advaage of he approach is ha re-rakig is carried ou eirely o he clie side ad he privacy of user ca be proeced. Xiag e al. [29] have proposed coex aware rakig. For example, if he user has clicked a URL i he previous search i he same sessio, he i is very likely she will o click he same URL agai whe i appears i he resul of he curre search. Tha meas ha ideally he URL should be demoed i he curre search resul. Probabilisic models for rakig have bee sudied i s- aisics ad relaed fields from may years ago. The mos popular oes iclude Placke-Luce model [20, 15], Mallows model [16], ad Bradley-Terry model [2, 10]. Placke-Luce model is a sagewise geeraive model, which decomposes he process of geeraig a permuaio of objecs io sequeial sages. Mallows model is a disace-based model, which defies he probabiliy of a permuaio accordig o is disace o a ceroid permuaio. Bradley-Terry model calculaes he probabiliy of a permuaio by pairwise comparisos. See [17] for a review o he opic. The Placke- Luce model has bee uilized i learig o rak [5], ad he Bradley-Terry model is uilized i rakig opimizaio i his paper. 3. RANKING OPTIMIZATION WITH CONSTRAINTS Le us describe he process of pos rakig. Give a query, a rakig lis of documes is firs creaed by he basic rakig model, presumably buil by learig o rak. Pos rakig may be he coduced, depedig o he query, user, or coex, which meas a refieme of he origial rakig lis, i which some documes are moved up ad some are moved dow. The origial rakig lis is creaed maily from he viewpoi of relevace bewee query ad documes. The refied rakig lis is furher creaed from he viewpoi of qualiy, diversiy, persoalizaio, coexualizaio, ad so o. The acios of pos processig ca be realized by usig heurisic rules, which is a commo approach i pracice. A rule ca be if he query is i a lis of ermiologies, he always have he Wikipedia page of he ermiology o he op oe posiio. Aoher rule ca be if he documes are rerieved by boh he origial query ad refied queries, he have a leas oe docume rerieved by he origial query raked a he op hree posiios (o reduce he risk of opic shif). Ye aoher rule ca be if he query is presideial elecio debaes, he make all he op webpages of differe rouds of he debaes grouped ogeher i he rakig lis. How o use rules for pos rakig is o a rivial issue. Firs, he rules may o be hard rules ad hey oly represe a guidelie for refiig he iiial rakig lis. For example, a rule may be he docume should be raked a op hree posiios, i which o specific posiio is decided. Secod, muliple rules migh be applied a he same ime, ad he rules migh be coradicory o each oher. Third, differe orders of applicaios of rules migh yield differe fial rakig resuls ad hus he order of applicaios eeds also be cosidered. Fially, he rules usually oly affec a

3 Figure 1: Illusraio of rakig opimizaio wih cosrais. small umber of documes, i is impora o make a balace bewee applicaio of rules ad preservaio of he origial rakig lis. I his paper, we formalize pos rakig as cosraied opimizaio, referred o as rakig opimizaio wih cosrais. The cosrais represe he rules for pos rakig, defied as fucios over ses of permuaios. The objecive fucio represes he rade-off bewee adherece o he origial rakig lis ad saisfacio of he cosrais. The cosrais are i fac sof cosrais i he formulaio. Pos rakig is aurally performed by coducig he opimizaio problem. Therefore, he issues described above ca all be aurally solved i he framework. Suppose ha σ deoes he origial rakig lis, C deoes he se of cosrais, ad π deoes he fial rakig lis of pos rakig. The opimizaio ca be wrie as follows mi L(σ, π) + λ R(C, π), (1) π Ω N where L deoes agreeme bewee σ ad π, R deoes saisficaio of C by π, λ deoes he radeoff coefficie, ad Ω N deoes he se of all permuaios (rakig liss) o he N documes i he curre search. I he opimizaio process, we eed o fid he opimal fial rakig π. Suppose he se of cosrais C is defied as C = {c i ( )}, c i : Ω N {0, 1}. If c i (π) = 1, he permuaio π violaes cosrai c i, oherwise, c i(π) = 0. The subse of permuaios which do o violae he cosrais are good cadidaes for he fial rakig lis. We ca defie several ypes of cosrais. Top-k cosrai: A docume mus be a op k posiios. No-op-k cosrai: A docume cao be a op k posiios. Cluserig cosrai: Two or more documes should be raked ogeher. Diversiy cosrai: Two or more documes should o be raked ogeher. Figure 1 illusraes he problem of rakig opimizaio wih cosrais (oe ha he figure is oly for illusraio purposes; he se of permuaios forms a discree se, o a Euclidea space.). Ω N is he se of all possible permuaios for N documes. Ω C is he subse of permuaios saisfyig he cosrais i C. σ is he origial permuaio give by he basic rakig model. I rakig opimizaio, we aim o fid he opimal permuaio (rakig lis) π which is as close o σ as possible ad i he meaime as wihi Ω C as possible. 4. OUR METHOD I his secio, we propose a mehod of rakig opimizaio o he basis of he Bradley-Terry model. The mehod oly makes use of he op k cosrai ad he o op k cosrai. We leave o fuure work he sudy of addig oher cosrais o he mehod. 4.1 Probabilisic Approach We cosider a probabilisic approach o rakig opimizaio. We assume ha here exiss a probabilisic rakig model M which gives rise o he rakig lis π, by π = arg max τ P (τ M) ad we icorporae M io he opimizaio problem (1) o obai mi L(σ, π, M) + λ R(C, π, M). π Ω N,M Tha is o say, we ur he opimizaio problem (1) wih respec o π io a opimizaio problem wih respec o boh π ad M. Le L(σ, π, M) = log P (σ M) ad R(C, π, M) = log P (C M), where P (σ M) is he probabiliy of geeraig he permuaio σ give M ad P (C M) is he probabiliy of geeraig all of he cosrais i C give M. We firs solve ad he solve mi log P (σ M) λ log P (C M), (2) M π = arg max π Ω N P (π M). (3) where π deoes he opimal rakig lis. The ierpreaio of he mehod is as follows. Give he rakig lis σ by he basic rakig model as well as he se of cosrais C, we wa o firs fid a probabiliy model M ha ca bes explai he rakig lis as well as he cosrais (i.e., he produc of he probabiliies P (σ M) ad P (C M) is he larges, wih a rade-off coefficie). Afer M is deermied, we wa o fid he bes rakig lis give by M (i.e., he probabiliy P (π M) is he larges). I his paper, we choose he Bradley-Terry model for calculaio of P (σ M) ad P (C M). 4.2 Usig Bradley-Terry Model The Bradley-Terry model represes he probabiliy disribuio of permuaio of N documes (i geeral iems) by makig compariso amog all pairs of documes. I assumes ha he rakig model M is parameerized wih a se of N scores Θ = (θ 1,, θ N ), each correspodig o a docume. Furhermore, he parameers are assumed o be posiive ad sum o oe, i.e., θ i > 0 for i = 1,, N ad N i=1 θ i = 1. I Bradley-Terry model, he probabiliy of a preferece pair (i, j) (docume i be raked higher ha j) is defied as p ij = P {(i, j)} = θ i. θ i + θ j Give he permuaio σ, he Bradley-Terry model defies he probabiliy P (σ M) as beig proporioal o he produc of probabiliies of i raked higher ha j for all preferece pairs (i, j): P (σ M) (i,j):σ(i)<σ(j) p ij = (i,j):σ(i)<σ(j) θ i θ i + θ j. Give he cosrai se C, he Bradley-Terry model defies he probabiliy P (C M) based o he preferece pairs derived from C: P (C M) θ i p ij =, θ i + θ j c C (i,j) P c c C (i,j) P c

4 where P c is he se of preferece pairs derived from he cosrai c. The mehods for derivig preferece pairs deped o he ypes of cosrais. Here, we give mehods for he opk cosrai ad o-op-k cosrai. Give a op-k cosrai c, he se of preferece pairs P c is defied as P c = {(i, j) j : σ(j) > k}, where i is he docume o promoe i cosrai c. Similarly, Give a o-op-k cosrai c, he se of preferece pairs is P c = {(j, i) j : σ(j) k}, where i is he docume o demoe i cosrai c. We oe ha he op-k ad o-op-k cosrais ca oly be defied o he origial rakig lis σ, because oly afer he rakig lis is give oe ca perform pos rakig o i, i.e., impose cosrais o i. Thus, he op-k ad o-op-k cosrais resric he posiios of a docume i he origial rakig σ. Thus, he opimizaio i Equaio (2) becomes mi Θ θ i f(θ) = log ρ c θ i log θ i + θ j θ i + θ j (i,j):σ(i)<σ(j) c C (i,j) P c (4) N subjec o i :θ i > 0, θ i = 1, i=1 where ρ c > 0 is he weigh for cosrai c. Noe ha parameer λ i Equaio (2) has bee merged o parameers ρ c s i Equaio (4). The fial rakig π, he, ca be obaied via he maximizaio i Equaio (3). Wih he use of Bradley-Terry model, i ca be simplified o sorig of he documes i descedig order of scores i Θ. 4.3 Opimizaio I is easy o demosrae ha f(θ) = f(α Θ) for ay θ α > 0 (oe ha log i αθ θ i +θ j = log i αθ i +αθ j ). Thus we ca ge rid of he cosrai N i=1 θ i = 1 i Equaio (4). This is because for ay soluio we ca ormalize he θ i s by dividig hem wih N i=1 θi. The ormalized soluio will saisfy he cosrai ad keep f uchaged. Furhermore, he cosrais of θ i > 0 ca be discarded by replacig θ i wih e s i for i = 1,, N. Thus, he opimizaio problem i (4) becomes he followig ucosraied opimizaio problem: mi S f(s) = (i,j):σ(i)<σ(j) (log(e s i + e s j ) s i ) + ρ c (log(e si + e sj ) s i), c C (i,j) P c where S = {s 1, s 2,, s N } is he se of parameers. The objecive fucio f(s) i Equaio (5) is covex, as saed i Theorem 4.1. Theorem 4.1. f(s) is a covex fucio. Proof of he heorem ca be foud i Appedix. Theorem 4.1 idicaes ha he objecive fucio f ca be efficiely opimized wih gradie desce. The gradie w.r.. S ca (5) Algorihm 1 Rakig Opimizaio Algorihm Require: Iiial rakig σ, cosrais C, ad shrikage rae 0 < α < 1 1: S (0) radom values 2: 1 3: repea 4: S = f S S=S ( 1) {Equaio (6)} 5: γ 1 {search opimal sep size usig backrackig} 6: while f(s ( 1) γ S) > f(s ( 1) ) γ S 2 2 do 7: γ αγ 8: ed while 9: S () S ( 1) γ S {Equaio (7)} 10: : uil covergece 12: reur Θ = { es 1 Z be wrie as f = { df S df = ds i + c C ρ c j:σ(j)<σ(i) j:(j,i) P c,, es N Z ds 1,, }, where Z = N =1 es df ds N }, where df ds i e s i e s i + e s j j:σ(i)<σ(j) s e i e s i + e s j j:(i,j) P c e s j is defied as e s i + e s j s e j e s i + e s j. Thus, he updaig crierio for gradie desce is S () = S ( 1) γ () f S, (7) S=S 1 where is he ieraio umber, γ () is he opimal sep size a he i-h ieraio which is deermied by backrackig. Algorihm 1 shows he pseudo code of he opimizaio algorihm. The fial docume rakig π, he, is obaied by sorig wih he scores i Θ reured i Algorihm Aalysis Covergece We aalyze he covergecy of Algorihm 1 ad have he followig heorem: Theorem 4.2. Algorihm 1 coverges i fiie seps ad he covergece rae is O( 1 ), where ϵ > 0 is he olerace. ϵ Proof of he heorem ca be foud i Appedix. Theorem 4.2 implies ha Algorihm 1 ca reur a opimal rakig model i a reasoably shor ime, which makes i possible o apply he algorihm olie Iuiive Explaaio The gradie i Equaio (6) has a iuiive explaaio. The gradie cosiss of wo pars, oe based o he origial rakig lis σ ad he oher based o he cosrai se C. Boh pars coribue o he gradie ad are calculaed o he basis of preferece pairs. Give a preferece pair (i, j), here will be a force ha pushes he preferred docume i upward, by subracig a posiive erm es j e s i +e s j = 1 e s i sj +1 o he gradie df ds i (6) (oe ha i gradie desce s i is updaed wih egaive gradie). The erm also idicaes ha he force is relaed o he

5 j Figure 2: Iuiive explaaio of he gradie. Give a preferece pair (i, j), docume i will be pushed upward (gree solid arrow) ad docume j will be pushed dowward (red dashed arrow). The sreghs of he forces are ideical. Table 1: Saisics of daases. daase # queries #documes #relevace levels MQ MQ OHSUMED Gov Eerprise differece bewee he scores. A small s i s j, which meas he curre scores s i ad s j do o agree wih he preferece pair, leads o a srog force o promoe docume i. I he same ime, he u-preferred docume j will be pushed dowward, by addig he same erm o gradie df ds j. Figure 2 illusraes he forces ha respecively push he docume i ad docume j upward ad dowward. Give all of he preferece pairs derived from σ ad from C, he overall forces ha promoe (or demoe) a docume i are joily deermied by all he preferece pairs relaed o docume i. 5. EXPERIMENTS We coduced experimes o es he performaces of our mehod for pos rakig (rakig opimizaio). 5.1 Experime Seig We kow of o exisece of public daa available for pos rakig. As approximaio, we used relevace daases for he experimes. Tha is, we assume ha we kow ha some documes are releva (i.e., should be raked high) ad boos he raks of he documes i pos rakig. We made use of he followig subses of he LETOR bechmark daase: MQ2007, MQ2008, OHSUMED ad.gov, as well as a daase of eerprise search, deoed as Eerprise. Table 1 gives he saisics of he four LETOR daases [14, 22] ad he Eerprise daase. The Eerprise daase cosiss of 183 queries; each query is associaed wih abou 30 documes. I oal, he daase coais 5464 query-docume pairs. Each query-docume pair is assiged wih a label represeig relevace a hree levels: Good, Fair, or Bad. The daase is spli io raiig daa (130 queries) ad es daa (53 queries). The rakig models were raied usig [28], which is a sae-of-he-ar mehod i learig o rak. The sadard feaures i LETOR daases were uilized ad we also defied 18 feaures for Eerprise, icludig BM25 [24] ad word level edi disace 1, ec. Two ypes of cosrais were esed i our experimes: he op-k cosrais ad o-op-k cosrais. For each es query, we geeraed oe op-k (k = 1, 3, 5) cosrai ad oe o-op-k (k = 5, 10) cosrai based o he labels 1 hp://e.wikipedia.org/wiki/edi disace i of documes wih respec o he query 2. Specifically, we sored he documes accordig o heir labels (o obai a perfec rakig) ad radomly seleced oe docume i from he op k posiios. The we creaed a cosrai c which saes ha he seleced docume i should be raked o op k posiios i he fial rakig. Similarly, we also creaed a o-op-k cosrai by radomly selec a docume j from posiios afer k i he perfec rakig. For simpliciy, we assumed ha all op-k (ad all o-opk) cosrais are equally impora ad ake he values of ρ (ad ρ ). Thus, he rakig opimizaio of Equaio (5) becomes mi f(s) = (log(e s i + e s j ) s i) S (i,j):σ(i)<σ(j) +ρ (i,j) P (log(e si + e sj ) s i) +ρ (i,j) P (log(e si + e sj ) s i ). As for baselie mehods, we use he followig four heurisics for modifyig he origial rakig: For he op-k cosrai, always raks he seleced docume o op oe posiio. For he oop-k cosrai, i always raks he seleced docume o he boom posiio of he lis. For he op-k cosrai, always raks he seleced docume o he middle of he op k posiios. For he o-op-k cosrai, i always raks he seleced docume o he middle of he remaiig lis afer k. For he op-k cosrai, always raks he docume o he posiio of k. For he o-op-k cosrai, always raks he docume o he posiio of k + 1. The above hree baselies do o cosider he posiio of he docume i he origial lis. promoes he docume seleced by he op-k cosrai o he posiio of k pos, where pos is he N rak of he docume i he origial rakig lis ad is he ceilig fucio. Therefore, he docume will be raked higher if is origial posiio is also high. Similarly, he baselie demoes he docume s- eleced by he o-op-k cosrai o he posiio of k + pos(1 k ). N As evaluaio measures, Normalized Discoued Cumulaive Gai (NDCG) [11] a posiios 1, 3, ad 5 were used. 5.2 Experimeal Resuls Resuls o LETOR I all he four daases i LETOR, he queries ad associaed documes were spli o 5 subses, ad 5-fold cross validaios were coduced. The performaces repored here are he averages over 5 rials. For each daase, we used he raiig daa o lear he basic rakig model, he validaio daa o ue he parameers, ad he es daa o perform rakig opimizaio. There are wo parameers ρ ad ρ ued wih he validae 2 Rakig opimizaio ca be coduced or o coduced depedig o queries. Here for experimeaio purpose, rakig opimizaio is assumed o be carried ou for all queries. (8)

6 (a) op-3, o-op-5, ρ =10, ρ =0 (b) op-3, o-op-10, ρ =10, ρ =10 (c) op-5, o-op-10, ρ =10, ρ =10 Figure 3: Rakig accuracies o MQ2008 daase. (a) op-3, o-op-5, ρ =100, ρ =10 (b) op-3, o-op-10, ρ =100, ρ =10 (c) op-5, o-op-10, ρ =100, ρ =10 Figure 4: Rakig accuracies o MQ2007 daase (a) op-3, o-op-5, ρ =200, ρ =0 (b) op-3, o-op-10, ρ =200, ρ =10 (c) op-5, o-op-10, ρ =200, ρ =30 Figure 5: Rakig accuracies o OHSUMED daase. (a) op-3, o-op-5, ρ =50, ρ =50 (b) op-3, o-op-10, ρ =50, ρ =50 (c) op-5, o-op-10, ρ =50, ρ =50 Figure 6: Rakig accuracies o.gov daase.

7 1 1 1 (a) op-3, o-op-5, ρ =60, ρ =40 (b) op-3, o-op-10, ρ =60, ρ =40 (c) op-5, o-op-10, ρ =60, ρ =40 Figure 7: Rakig accuracies o Eerprise daase. Table 2: Average ime (i millisecods) of rakig opimizaio i seig of (op-5, o-op-10). MQ2008 MQ2007 OHSUMED.Gov Eerprise ime se. The performaces o he es ses are hose based o he bes performig parameers. We esed all he six combiaios of op-k cosrai (k = 1, 3, 5) ad o-op-k cosrai (k = 5, 10) o all of he four daases. Figures 3, 4, 5, ad 6 repor he experimeal resuls o MQ2008, MQ2007, OHSUMED, ad.gov, respecively. Our rakig opimizaio mehod is deoed as i he figures. The bes performig parameers of ρ ad ρ are also show. The performaces of he combiaios (op-3, o-op-5), (op-3, o-op-10), ad (op-5, o-op-10) are repored. From he resuls, we ca see ha our mehod ouperforms he baselies as well as he basic rakig model of o all daases. The experimeal resuls for (op- 3, o-op-5) ad (op-3, o-op-10) are very similar. This is because all he mehods focus o he op of rakig lis. The chages of k o he o-op-k cosrai have less impac o he fial rakig lis. Our mehod ad he baselies perform equally well, whe k is 1 for he op-k cosrai. This is because he op-1 cosrai meas promoig he docume o he op oe posiio, ad our mehod ca fucio as a hard rule i such case ad produce he same rakig resul as he baselies (we omi he resul because of space limiaio). We will have discussios o he effec of differe k values i Secio 5.3. We coduced sigifica ess (-es) o he improvemes of our mehod over all he baselies. The resuls idicae ha all he improvemes are saisically sigifica (p-value < 0.05), excep (op-5, o-op-10) over o MQ2008 i erms of NDCG@5, (op-3, o-op-5) ad (op-3, o-op-10) over o OHSUMED ad.gov i erms of NDCG@3 ad NDCG@5. Noe ha whe op-1 cosrai is adoped, our mehod performs equally well wih he baselies. Table 2 repors average ruig ime of rakig opimizaio by our mehod (wih op-5 ad o-op-10 cosrais) i millisecods o a Lapop PC wih 2.4GHZ CPU ad 4G- B memory. We ca see ha our mehod rus very fas, eve o a upowerful machie. We also observed ha for mos queries he algorihm coverges wihi 10 ieraios. Similar experimeal resuls were also observed i oher experimes. The resuls empirically verify he coclusio of Theorem Resuls o Eerprise I he experime, we uilized he raiig daa o lear he basic rakig model ad o ue he parameers ρ ad ρ. The es daa was uilized o perform rakig opimizaio. We esed all he six combiaios of op-k cosrai (k = 1, 3, 5) ad o-op-k cosrai (k = 5, 10) o he Eerprise daase. The performaces of he combiaios (op-3, o-op-5), (op-3, o-op-10), ad (op-5, oop-10) are repored i Figure 8. Our rakig opimizaio mehod is deoed as i he figures. From he resuls, we ca see ha our mehod ouperforms he baselies as well as he basic rakig model of o all daases. The experimeal resuls for (op-3, o-op-5) ad (op-3, o-op-10) are very similar, as i he experimes o LETOR daases. We coduced -es o he improvemes of our mehod over he baselies i erms of NDCG@1, NDCG@3 ad NDCG@5. The improvemes are saisically sigifica. Agai, our mehod ad he baselies perform equally well whe he op-1 cosrai is adoped. We omi he resul because of space limiaio. We also esed he ruig ime of our mehod of rakig opimizaio, i he seig of op-5 ad o-op-10 cosrais. The las colum of Table 2 repors average ime of our mehod i millisecods. We ca see ha our mehod rus very fas. Similar experimeal resuls were obaied for oher combiaios of op-k ad o-op-k cosrais. 5.3 Discussios We firs ivesigaed he reasos ha our mehod of rakig opimizaio ouperforms he baselie mehods. We foud ha i geeral our mehod ca really work beer ha he baselies o make a good comprise bewee usig he origial rakig ad usig he cosrais. Here, we use he resul of MQ2008 daase wih regard o wo queries o illusrae why our mehod is superior o he baselies. Figure 8(a) shows he origial rakig by ad he fial rakigs by differe mehods wih regard o oe query. The empy blocks, grid blocks, ad filled blocks represe o releva documes, parially releva documes, ad releva documes, respecively. The documes seleced by op-k cosrais ad o-op-k cosrais are marked wih ad, respecively. From he resuls, we ca see ha our mehod of rakig opimiza-

8 5 5 5 (a) Example 1 (b) Example k (a) Top-k cosrai Figure 8: Example rakigs from MQ2008. io ca really promoe he releva docume (marked wih ) ad demoe he o releva documes (marked wih ). I his case, our mehod ouperforms he baselies of,, ad. The resuls wih regard o he oher query are repored i Figure 8(b), which is a oisy case i which he docume seleced by he op-k cosrai is acually o releva (oises may also exis i he rules i pracice). Our mehod of rakig opimizaio also akes io accou he posiio of he docume i he origial rakig lis give by ad hus does o promoe i oo much. I oher words, our mehod ca make a good rade-off bewee he cosrais ad he origial rakig. O he oher had, he baselies of ad do o have such cosideraio ad cao rak he docume saisfacorily. Therefore, our mehod is more capable for pos rakig ha he baselies. We furher coduced experimes o see he impac of differe k values o differe mehods. For he op-k cosrai, larger k meas a sofer corol o he fial rakig lis. Whe k = 1, he cosrai becomes a hard rule. From he resuls repored i Figure 9(a), we ca see ha our mehod of rakig opimizaio always works beer ha or as well as he baselies for all k values. Whe k ges close o oe, our mehod will perform similarly as he baselies; whe k ges larger, he improvemes of our mehod over he baselies will also be larger. The resuls idicae ha our mehod is more robus ha he baselies. We also oe ha he baselie of hurs he basic rakig resuls whe k ges large. This is because is very sesiive o he oise i he cosrais (rules). The resuls idicae ha i is risky o direcly apply o pos rakig hough i ouperforms he oher baselies i mos of he experimes. O he oher had, for he o-op-k cosrai, smaller k meas a sofer corol o he fial rakig lis. Whe k = N 1, he cosrai becomes a hard rule. From he resuls repored i Figure 9(b), we ca see ha our mehod of rakig opimizaio always works beer ha or as well as he baselies for all k values. Whe k ges close o oe, our mehod will perform similarly as he baselies, because he o-op-k cosrai has a sof corol o he fial rakig lis. Whe k ges larger, he improvemes of our mehod over he baselies will also become larger. There is a peak k (b) No-op-k cosrai Figure 9: Performaces of rakig opimizaio wih respec o differe k values i erms of NDCG@1. for performace of our mehod aroud k = 5. The performace will drop afer k = 5. This is because he o-op-k cosrai impacs more o he ail of he rakig lis, ad a larger k will have less impac o he op of he rakig lis, which is more impora i rakig evaluaio. We also ivesigaed he ifluece of differe ypes of cosrais. Specifically, we esed performaces of our mehod of rak opimizaio wihou he cosrais ( oly), wih op-5 cosrai oly, wih o-op-10 cosrai oly, ad wih boh ypes of cosrais. Figure 10 repors he resuls. From he figure we ca see ha he wo ypes of cosrais ca idividually improve he rakig performaces if hey are adoped. The performaces ca be furher improved if hey are used simulaeously. The resuls idicae ha our mehod of rakig opimizaio ca leverage muliple ypes of cosrais wihi oe framework. The op-5 cosrai ouperforms he o-op-10 cosrai, because he evaluaio measure of NDCG emphasizes he imporace of op rakig, which is also he focus of he op-k-cosrai. Fially, we evaluaed how sesiive our mehod of rakig opimizaio is o he parameer seigs. I he experime, wo parameers ρ ad ρ were esed, which are weighs of he op-k cosrai ad o-op-k cosrai, respecively.

9 5 5 5 Oly op k Oly o op k Boh Figure 10: Performaces of rakig opimizaio wih differe cosrai ypes. NDCG NDCG NDCG@1 NDCG@3 NDCG@ weigh (a) ρ NDCG@1 NDCG@3 NDCG@ weigh (b) ρ Figure 11: Performaces of rakig opimizaio wih respec o differe parameer seigs. We chaged oe parameer ad fixed he oher o is opimal value. Figure 11(a) ad Figure 11(b) show he performaces of rakig opimizaio of our mehod i erms of NDCG a he posiios of 1, 3, ad 5. From he resuls, we ca see ha our mehod is o sesiive o he parameer seigs, ad hus is quie robus. 6. CONCLUSION I his paper, we have sudied he problem of pos processig of rakig i search, which we call pos rakig. Pracices o pos rakig i realiy ed o be heurisic ad we have, perhaps for he firs ime, formalized he problem as a opimizaio problem. I he formulaio, we represe he pos processig rules as cosrais ad maage o miimize he objecive fucio deoig he rade-off bewee he origial rakig ad he cosrais. As a resul, oe ca perform pos rakig hrough solvig he opimizaio issue. We have also give a specific probabilisic implemeaio of he opimizaio formulaio. Barley-Terry model is employed o calculaig he probabiliy of rakig lis. The objecive fucio is defied based o he codiioal probabiliy of he origial rakig ad he codiioal probabiliy of he cosrais give he model. The opimizaio amous o miimizaio of he sum of he egaive log probabiliies. We have compared he performaces of our mehod wih several baselies i experimes usig a umber of daases icludig bechmark daases. The baselie mehods represe pracical mehods of usig rules. The resuls show ha i is always beer o employ our mehod i pos rakig ha he baselies. There are sill may ope quesios wih regard o rakig opimizaio. We pla o coduc more research o he problem i he fuure. The ope quesios iclude (1) wheher here exiss a more geeral framework for rakig opimizaio, (2) how o defie ad icorporae oher ypes of cosrais io he framework, (3) how o aurally add diversificaio of resuls, ec. io he framework, (4) wheher here are more effecive ad efficie mehods for he ask. 7. REFERENCES [1] S. Boyd ad L. Vadeberghe. Covex Opimizaio. Cambridge Uiversiy Press, New York, USA, [2] R. A. Bradley ad M. E. Terry. The rak aalysis of icomplee block desigs I. The mehod of paired comparisos. Biomerika, 39: , [3] C. Burges, T. Shaked, E. Reshaw, A. Lazier, M. Deeds, N. Hamilo, ad G. Hulleder. Learig o rak usig gradie desce. I Proceedigs of he 22Nd Ieraioal Coferece o Machie Learig, ICML 05, pages 89 96, [4] Y. Cao, J. Xu, T.-Y. Liu, H. Li, Y. Huag, ad H.-W. Ho. Adapig rakig svm o docume rerieval. I Proceedigs of he 29h Aual Ieraioal ACM SIGIR Coferece, SIGIR 06, pages , [5] Z. Cao, T. Qi, T.-Y. Liu, M.-F. Tsai, ad H. Li. Learig o rak: from pairwise approach o liswise approach. I ICML 07: Proceedigs of he 24h ieraioal coferece o Machie learig, pages , [6] J. Carboell ad J. Goldsei. The use of mmr, diversiy-based rerakig for reorderig documes ad producig summaries. I Proceedigs of he 21s Aual Ieraioal ACM SIGIR Coferece, SIGIR 98, pages , [7] Z. Dou, S. Hu, K. Che, R. Sog, ad J.-R. We. Muli-dimesioal search resul diversificaio. I

10 Proceedigs of he Fourh ACM Ieraioal Coferece o Web Search ad Daa Miig, WSDM 11, pages , [8] Y. Freud, R. Iyer, R. E. Schapire, ad Y. Siger. A efficie boosig algorihm for combiig prefereces. J. Mach. Lear. Res., 4: , Dec [9] R. Herbrich, T. Graepel, ad K. Obermayer. Large margi rak boudaries for ordial regressio. I Advaces i Large Margi Classifiers, pages , Cambridge, MA, MIT Press. [10] R. Huer. Mm algorihms for geeralized bradley-erry models. The Aals of Saisics, 32:2004, [11] K. Järveli ad J. Kekäläie. Cumulaed gai-based evaluaio of ir echiques. ACM Tras. If. Sys., 20(4): , Oc [12] J. Laffery ad C. Zhai. Docume laguage models, query models, ad risk miimizaio for iformaio rerieval. I Proceedigs of he 24h Aual Ieraioal ACM SIGIR Coferece, SIGIR 01, pages , [13] P. Li, C. J. C. Burges, ad Q. Wu. Mcrak: Learig o rak usig muliple classificaio ad gradie boosig. I NIPS, [14] T.-Y. Liu, J. Xu, T. Qi, W. Xiog, ad H. Li. Leor: Bechmark daase for research o learig o rak for iformaio rerieval. I Proceedigs of SIGIR 2007 workshop o learig o rak for iformaio rerieval, pages 3 10, [15] R. D. Luce. Idividual Choice Behavior: A heoreical aalysis. Wiley, [16] C. L. Mallows. No-ull rakig models. i. Biomerika, 44(1-2): , Jue [17] J. I. Marde. Aalyzig ad Modelig Rak Daa. Chapma & Hall, [18] R. Nallapai. Discrimiaive models for iformaio rerieval. I Proceedigs of he 27h Aual Ieraioal ACM SIGIR Coferece, SIGIR 04, pages 64 71, [19] Y. Neserov ad I. Neserov. Iroducory Lecures o Covex Opimizaio: A Basic Course. Applied Opimizaio. Spriger, [20] R. L. Placke. The aalysis of permuaios. Applied Saisics, 24: , [21] J. M. Poe ad W. B. Crof. A laguage modelig approach o iformaio rerieval. I Proceedigs of he 21s Aual Ieraioal ACM SIGIR Coferece, SIGIR 98, pages , [22] T. Qi, T.-Y. Liu, J. Xu, ad H. Li. Leor: A bechmark collecio for research o learig o rak for iformaio rerieval. If. Rer., 13(4): , Aug [23] F. Radliski ad S. Dumais. Improvig persoalized web search usig resul diversificaio. I Proceedigs of he 29h Aual Ieraioal ACM SIGIR Coferece, SIGIR 06, pages , [24] S. Roberso ad D. A. Hull. The rec-9 filerig rack fial repor. pages 25 40, [25] K. Sugiyama, K. Haao, ad M. Yoshikawa. Adapive web search based o user profile cosruced wihou ay effor from users. I Proceedigs of he 13h Ieraioal Coferece o World Wide Web, pages , [26] J. Teeva, S. T. Dumais, ad E. Horviz. Persoalizig search via auomaed aalysis of ieress ad aciviies. I Proceedigs of he 28h Aual Ieraioal ACM SIGIR Coferece, SIGIR 05, pages , [27] E. Vee, U. Srivasava, J. Shamugasudaram, P. Bha, ad S. Yahia. Efficie compuaio of diverse query resuls. I IEEE 24h Ieraioal Coferece o Daa Egieerig, pages , [28] Q. Wu, C. J. Burges, K. M. Svore, ad J. Gao. Adapig boosig for iformaio rerieval measures. If. Rer., 13(3): , Jue [29] B. Xiag, D. Jiag, J. Pei, X. Su, E. Che, ad H. Li. Coex-aware rakig i web search. I Proceedigs of he 33rd Ieraioal ACM SIGIR Coferece, SIGIR 10, pages , [30] J. Xu ad H. Li. Adarak: A boosig algorihm for iformaio rerieval. I Proceedigs of he 30h Aual Ieraioal ACM SIGIR Coferece, SIGIR 07, pages , [31] J. Xu, T.-Y. Liu, M. Lu, H. Li, ad W.-Y. Ma. Direcly opimizig evaluaio measures i learig o rak. I Proceedigs of he 31s Aual Ieraioal ACM SIGIR Coferece, SIGIR 08, pages , [32] Y. Yue, T. Filey, F. Radliski, ad T. Joachims. A suppor vecor mehod for opimizig average precisio. I Proceedigs of he 30h Aual Ieraioal ACM SIGIR Coferece, SIGIR 07, pages , Appedix Proof of Theorem 4.1. Proof. The objecive fucio f ca be decomposed as f = f a f b, where f a = (i,j) P a (i,j) log(e s i + e s j ) ad f b = N i=1 b i s i. Here a (i,j) 0 for all (i, j) pairs i P, which is he uio of all preferece pairs. b i 0 for i = 1,, N. e s i is a log-covex fucio w.r. s i. Thus, log(e s i + e s j ) is also covex for ay (i, j) P, because log-covexiy ca be persevered uder addiio (cf., [1]). Hece, f a is covex because a (i,j) 0 holds for all (i, j) P. Fially, we coclude ha f = f a f b is covex because f b is liear. Proof of Theorem 4.2. Proof. Sice s i < log(e s i + e s j ) holds for ay preferece pair (i, j), obviously he objecive fucio f i Equaio (5) has a lower boud of 0. Also, accordig o Theorem 4.1, f is covex. Thus, here exis oe ad oly oe opimal soluio for he problem of Equaio (5), deoed as S. Sice gradie desce ad backrackig are adoped for he opimizaio, he followig iequaliy holds (cf., [19]): f(s () ) f(s ) S(0) S 2, 2η mi where is he umber of ieraios, η mi = mi{1, α/l}, L is Lipschiz cosa for f, ad α is he shrikage rae. Thus, Algorihm 1 ca coverge wihi O( 1 ) o reach he ϵ olerace ϵ.

FORECASTING MODEL FOR AUTOMOBILE SALES IN THAILAND

FORECASTING MODEL FOR AUTOMOBILE SALES IN THAILAND FORECASTING MODEL FOR AUTOMOBILE SALES IN THAILAND by Wachareepor Chaimogkol Naioal Isiue of Developme Admiisraio, Bagkok, Thailad Email: wachare@as.ida.ac.h ad Chuaip Tasahi Kig Mogku's Isiue of Techology

More information

CHAPTER 22 ASSET BASED FINANCING: LEASE, HIRE PURCHASE AND PROJECT FINANCING

CHAPTER 22 ASSET BASED FINANCING: LEASE, HIRE PURCHASE AND PROJECT FINANCING CHAPTER 22 ASSET BASED FINANCING: LEASE, HIRE PURCHASE AND PROJECT FINANCING Q.1 Defie a lease. How does i differ from a hire purchase ad isalme sale? Wha are he cash flow cosequeces of a lease? Illusrae.

More information

Bullwhip Effect Measure When Supply Chain Demand is Forecasting

Bullwhip Effect Measure When Supply Chain Demand is Forecasting J. Basic. Appl. Sci. Res., (4)47-43, 01 01, TexRoad Publicaio ISSN 090-4304 Joural of Basic ad Applied Scieific Research www.exroad.com Bullwhip Effec Measure Whe Supply Chai emad is Forecasig Ayub Rahimzadeh

More information

Ranking of mutually exclusive investment projects how cash flow differences can solve the ranking problem

Ranking of mutually exclusive investment projects how cash flow differences can solve the ranking problem Chrisia Kalhoefer (Egyp) Ivesme Maageme ad Fiacial Iovaios, Volume 7, Issue 2, 2 Rakig of muually exclusive ivesme projecs how cash flow differeces ca solve he rakig problem bsrac The discussio abou he

More information

A panel data approach for fashion sales forecasting

A panel data approach for fashion sales forecasting A pael daa approach for fashio sales forecasig Shuyu Re(shuyu_shara@live.c), Tsa-Mig Choi, Na Liu Busiess Divisio, Isiue of Texiles ad Clohig, The Hog Kog Polyechic Uiversiy, Hug Hom, Kowloo, Hog Kog Absrac:

More information

A Queuing Model of the N-design Multi-skill Call Center with Impatient Customers

A Queuing Model of the N-design Multi-skill Call Center with Impatient Customers Ieraioal Joural of u- ad e- ervice, ciece ad Techology Vol.8, o., pp.- hp://dx.doi.org/./ijuess..8.. A Queuig Model of he -desig Muli-skill Call Ceer wih Impaie Cusomers Chuya Li, ad Deua Yue Yasha Uiversiy,

More information

REVISTA INVESTIGACION OPERACIONAL VOL. 31, No.2, 159-170, 2010

REVISTA INVESTIGACION OPERACIONAL VOL. 31, No.2, 159-170, 2010 REVISTA INVESTIGACION OPERACIONAL VOL. 3, No., 59-70, 00 AN ALGORITHM TO OBTAIN AN OPTIMAL STRATEGY FOR THE MARKOV DECISION PROCESSES, WITH PROBABILITY DISTRIBUTION FOR THE PLANNING HORIZON. Gouliois E.

More information

Managing Learning and Turnover in Employee Staffing*

Managing Learning and Turnover in Employee Staffing* Maagig Learig ad Turover i Employee Saffig* Yog-Pi Zhou Uiversiy of Washigo Busiess School Coauhor: Noah Gas, Wharo School, UPe * Suppored by Wharo Fiacial Isiuios Ceer ad he Sloa Foudaio Call Ceer Operaios

More information

Studies in sport sciences have addressed a wide

Studies in sport sciences have addressed a wide REVIEW ARTICLE TRENDS i Spor Scieces 014; 1(1: 19-5. ISSN 99-9590 The eed o repor effec size esimaes revisied. A overview of some recommeded measures of effec size MACIEJ TOMCZAK 1, EWA TOMCZAK Rece years

More information

Research Article Dynamic Pricing of a Web Service in an Advance Selling Environment

Research Article Dynamic Pricing of a Web Service in an Advance Selling Environment Hidawi Publishig Corporaio Mahemaical Problems i Egieerig Volume 215, Aricle ID 783149, 21 pages hp://dx.doi.org/1.1155/215/783149 Research Aricle Dyamic Pricig of a Web Service i a Advace Sellig Evirome

More information

The Term Structure of Interest Rates

The Term Structure of Interest Rates The Term Srucure of Ieres Raes Wha is i? The relaioship amog ieres raes over differe imehorizos, as viewed from oday, = 0. A cocep closely relaed o his: The Yield Curve Plos he effecive aual yield agais

More information

UNDERWRITING AND EXTRA RISKS IN LIFE INSURANCE Katarína Sakálová

UNDERWRITING AND EXTRA RISKS IN LIFE INSURANCE Katarína Sakálová The process of uderwriig UNDERWRITING AND EXTRA RISKS IN LIFE INSURANCE Kaaría Sakálová Uderwriig is he process by which a life isurace compay decides which people o accep for isurace ad o wha erms Life

More information

Distributed Containment Control with Multiple Dynamic Leaders for Double-Integrator Dynamics Using Only Position Measurements

Distributed Containment Control with Multiple Dynamic Leaders for Double-Integrator Dynamics Using Only Position Measurements IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 57, NO. 6, JUNE 22 553 Disribued Coaime Corol wih Muliple Dyamic Leaders for Double-Iegraor Dyamics Usig Oly Posiio Measuremes Jiazhe Li, Wei Re, Member, IEEE,

More information

Modeling the Nigerian Inflation Rates Using Periodogram and Fourier Series Analysis

Modeling the Nigerian Inflation Rates Using Periodogram and Fourier Series Analysis CBN Joural of Applied Saisics Vol. 4 No.2 (December, 2013) 51 Modelig he Nigeria Iflaio Raes Usig Periodogram ad Fourier Series Aalysis 1 Chukwuemeka O. Omekara, Emmauel J. Ekpeyog ad Michael P. Ekeree

More information

Kyoung-jae Kim * and Ingoo Han. Abstract

Kyoung-jae Kim * and Ingoo Han. Abstract Simulaeous opimizaio mehod of feaure rasformaio ad weighig for arificial eural eworks usig geeic algorihm : Applicaio o Korea sock marke Kyoug-jae Kim * ad Igoo Ha Absrac I his paper, we propose a ew hybrid

More information

Combining Adaptive Filtering and IF Flows to Detect DDoS Attacks within a Router

Combining Adaptive Filtering and IF Flows to Detect DDoS Attacks within a Router KSII RANSAIONS ON INERNE AN INFORMAION SYSEMS VOL. 4, NO. 3, Jue 2 428 opyrigh c 2 KSII ombiig Adapive Filerig ad IF Flows o eec os Aacks wihi a Rouer Ruoyu Ya,2, Qighua Zheg ad Haifei Li 3 eparme of ompuer

More information

1/22/2007 EECS 723 intro 2/3

1/22/2007 EECS 723 intro 2/3 1/22/2007 EES 723 iro 2/3 eraily, all elecrical egieers kow of liear sysems heory. Bu, i is helpful o firs review hese coceps o make sure ha we all udersad wha his heory is, why i works, ad how i is useful.

More information

Reaction Rates. Example. Chemical Kinetics. Chemical Kinetics Chapter 12. Example Concentration Data. Page 1

Reaction Rates. Example. Chemical Kinetics. Chemical Kinetics Chapter 12. Example Concentration Data. Page 1 Page Chemical Kieics Chaper O decomposiio i a isec O decomposiio caalyzed by MO Chemical Kieics I is o eough o udersad he soichiomery ad hermodyamics of a reacio; we also mus udersad he facors ha gover

More information

APPLICATIONS OF GEOMETRIC

APPLICATIONS OF GEOMETRIC APPLICATIONS OF GEOMETRIC SEQUENCES AND SERIES TO FINANCIAL MATHS The mos powerful force i he world is compoud ieres (Alber Eisei) Page of 52 Fiacial Mahs Coes Loas ad ivesmes - erms ad examples... 3 Derivaio

More information

Why we use compounding and discounting approaches

Why we use compounding and discounting approaches Comoudig, Discouig, ad ubiased Growh Raes Near Deb s school i Souher Colorado. A examle of slow growh. Coyrigh 000-04, Gary R. Evas. May be used for o-rofi isrucioal uroses oly wihou ermissio of he auhor.

More information

Mechanical Vibrations Chapter 4

Mechanical Vibrations Chapter 4 Mechaical Vibraios Chaper 4 Peer Aviabile Mechaical Egieerig Deparme Uiversiy of Massachuses Lowell 22.457 Mechaical Vibraios - Chaper 4 1 Dr. Peer Aviabile Modal Aalysis & Corols Laboraory Impulse Exciaio

More information

Chapter 8: Regression with Lagged Explanatory Variables

Chapter 8: Regression with Lagged Explanatory Variables Chaper 8: Regression wih Lagged Explanaory Variables Time series daa: Y for =1,..,T End goal: Regression model relaing a dependen variable o explanaory variables. Wih ime series new issues arise: 1. One

More information

THE FOREIGN EXCHANGE EXPOSURE OF CHINESE BANKS

THE FOREIGN EXCHANGE EXPOSURE OF CHINESE BANKS Workig Paper 07/2008 Jue 2008 THE FOREIGN ECHANGE EPOSURE OF CHINESE BANKS Prepared by Eric Wog, Jim Wog ad Phyllis Leug 1 Research Deparme Absrac Usig he Capial Marke Approach ad equiy-price daa of 14

More information

A Strategy for Trading the S&P 500 Futures Market

A Strategy for Trading the S&P 500 Futures Market 62 JOURNAL OF ECONOMICS AND FINANCE Volume 25 Number 1 Sprig 2001 A Sraegy for Tradig he S&P 500 Fuures Marke Edward Olszewski * Absrac A sysem for radig he S&P 500 fuures marke is proposed. The sysem

More information

Improving Survivability through Traffic Engineering in MPLS Networks

Improving Survivability through Traffic Engineering in MPLS Networks Improvig Survivabiiy hrough Traffic Egieerig i MPLS Neworks Mia Ami, Ki-Ho Ho, George Pavou, ad Michae Howarh Cere for Commuicaio Sysems Research, Uiversiy of Surrey, UK Emai:{M.Ami, K.Ho, G.Pavou, M.Howarh}@eim.surrey.ac.uk

More information

Financial Data Mining Using Genetic Algorithms Technique: Application to KOSPI 200

Financial Data Mining Using Genetic Algorithms Technique: Application to KOSPI 200 Fiacial Daa Miig Usig Geeic Algorihms Techique: Applicaio o KOSPI 200 Kyug-shik Shi *, Kyoug-jae Kim * ad Igoo Ha Absrac This sudy ieds o mie reasoable radig rules usig geeic algorihms for Korea Sock Price

More information

Modelling Time Series of Counts

Modelling Time Series of Counts Modellig ime Series of Cous Richard A. Davis Colorado Sae Uiversiy William Dusmuir Uiversiy of New Souh Wales Yig Wag Colorado Sae Uiversiy /3/00 Modellig ime Series of Cous wo ypes of Models for Poisso

More information

PERFORMANCE COMPARISON OF TIME SERIES DATA USING PREDICTIVE DATA MINING TECHNIQUES

PERFORMANCE COMPARISON OF TIME SERIES DATA USING PREDICTIVE DATA MINING TECHNIQUES , pp.-57-66. Available olie a hp://www.bioifo.i/coes.php?id=32 PERFORMANCE COMPARISON OF TIME SERIES DATA USING PREDICTIVE DATA MINING TECHNIQUES SAIGAL S. 1 * AND MEHROTRA D. 2 1Deparme of Compuer Sciece,

More information

A formulation for measuring the bullwhip effect with spreadsheets Una formulación para medir el efecto bullwhip con hojas de cálculo

A formulation for measuring the bullwhip effect with spreadsheets Una formulación para medir el efecto bullwhip con hojas de cálculo irecció y rgaizació 48 (01) 9-33 9 www.revisadyo.com A formulaio for measurig he bullwhip effec wih spreadshees Ua formulació para medir el efeco bullwhip co hojas de cálculo Javier Parra-Pea 1, Josefa

More information

Introduction to Statistical Analysis of Time Series Richard A. Davis Department of Statistics

Introduction to Statistical Analysis of Time Series Richard A. Davis Department of Statistics Iroduio o Saisial Aalysis of Time Series Rihard A. Davis Deparme of Saisis Oulie Modelig obeives i ime series Geeral feaures of eologial/eviromeal ime series Compoes of a ime series Frequey domai aalysis-he

More information

FEBRUARY 2015 STOXX CALCULATION GUIDE

FEBRUARY 2015 STOXX CALCULATION GUIDE FEBRUARY 2015 STOXX CALCULATION GUIDE STOXX CALCULATION GUIDE CONTENTS 2/23 6.2. INDICES IN EUR, USD AND OTHER CURRENCIES 10 1. INTRODUCTION TO THE STOXX INDEX GUIDES 3 2. CHANGES TO THE GUIDE BOOK 4 2.1.

More information

Task is a schedulable entity, i.e., a thread

Task is a schedulable entity, i.e., a thread Real-Time Scheduling Sysem Model Task is a schedulable eniy, i.e., a hread Time consrains of periodic ask T: - s: saring poin - e: processing ime of T - d: deadline of T - p: period of T Periodic ask T

More information

http://www.ejournalofscience.org Monitoring of Network Traffic based on Queuing Theory

http://www.ejournalofscience.org Monitoring of Network Traffic based on Queuing Theory VOL., NO., November ISSN XXXX-XXXX ARN Joural of Sciece a Techology - ARN Jourals. All righs reserve. hp://www.ejouralofsciece.org Moiorig of Newor Traffic base o Queuig Theory S. Saha Ray,. Sahoo Naioal

More information

4. Levered and Unlevered Cost of Capital. Tax Shield. Capital Structure

4. Levered and Unlevered Cost of Capital. Tax Shield. Capital Structure 4. Levered ad levered Cos Capial. ax hield. Capial rucure. Levered ad levered Cos Capial Levered compay ad CAP he cos equiy is equal o he reur expeced by sockholders. he cos equiy ca be compued usi he

More information

Single-machine Scheduling with Periodic Maintenance and both Preemptive and. Non-preemptive jobs in Remanufacturing System 1

Single-machine Scheduling with Periodic Maintenance and both Preemptive and. Non-preemptive jobs in Remanufacturing System 1 Absrac number: 05-0407 Single-machine Scheduling wih Periodic Mainenance and boh Preempive and Non-preempive jobs in Remanufacuring Sysem Liu Biyu hen Weida (School of Economics and Managemen Souheas Universiy

More information

DBIQ USD Investment Grade Corporate Bond Interest Rate Hedged Index

DBIQ USD Investment Grade Corporate Bond Interest Rate Hedged Index db Idex Developme Sepember 2014 DBIQ Idex Guide DBIQ USD Ivesme Grade Corporae Bod Ieres Rae Hedged Idex Summary The DBIQ USD Ivesme Grade Corporae Bod Ieres Rae Hedged Idex (he Idex ) is a rule based

More information

Derivative Securities: Lecture 7 Further applications of Black-Scholes and Arbitrage Pricing Theory. Sources: J. Hull Avellaneda and Laurence

Derivative Securities: Lecture 7 Further applications of Black-Scholes and Arbitrage Pricing Theory. Sources: J. Hull Avellaneda and Laurence Deivaive ecuiies: Lecue 7 uhe applicaios o Black-choles ad Abiage Picig heoy ouces: J. Hull Avellaeda ad Lauece Black s omula omeimes is easie o hik i ems o owad pices. Recallig ha i Black-choles imilaly

More information

14 Protecting Private Information in Online Social Networks

14 Protecting Private Information in Online Social Networks 4 roecig rivae Iormaio i Olie Social eworks Jiamig He ad Wesley W. Chu Compuer Sciece Deparme Uiversiy o Calioria USA {jmhekwwc}@cs.ucla.edu Absrac. Because persoal iormaio ca be ierred rom associaios

More information

A New Hybrid Network Traffic Prediction Method

A New Hybrid Network Traffic Prediction Method This full ex paper was peer reviewed a he direcio of IEEE Couicaios Sociey subjec aer expers for publicaio i he IEEE Globeco proceedigs. A New Hybrid Nework Traffic Predicio Mehod Li Xiag, Xiao-Hu Ge,

More information

Circularity and the Undervaluation of Privatised Companies

Circularity and the Undervaluation of Privatised Companies CMPO Workig Paper Series No. 1/39 Circulariy ad he Udervaluaio of Privaised Compaies Paul Grou 1 ad a Zalewska 2 1 Leverhulme Cere for Marke ad Public Orgaisaio, Uiversiy of Brisol 2 Limburg Isiue of Fiacial

More information

An Approach for Measurement of the Fair Value of Insurance Contracts by Sam Gutterman, David Rogers, Larry Rubin, David Scheinerman

An Approach for Measurement of the Fair Value of Insurance Contracts by Sam Gutterman, David Rogers, Larry Rubin, David Scheinerman A Approach for Measureme of he Fair Value of Isurace Coracs by Sam Guerma, David Rogers, Larry Rubi, David Scheierma Absrac The paper explores developmes hrough 2006 i he applicaio of marke-cosise coceps

More information

Optimal Combination of International and Inter-temporal Diversification of Disaster Risk: Role of Government. Tao YE, Muneta YOKOMATSU and Norio OKADA

Optimal Combination of International and Inter-temporal Diversification of Disaster Risk: Role of Government. Tao YE, Muneta YOKOMATSU and Norio OKADA 京 都 大 学 防 災 研 究 所 年 報 第 5 号 B 平 成 9 年 4 月 Auals of Disas. Prev. Res. Is., Kyoo Uiv., No. 5 B, 27 Opimal Combiaio of Ieraioal a Ier-emporal Diversificaio of Disaser Risk: Role of Goverme Tao YE, Muea YOKOMATSUaNorio

More information

Chapter 4 Return and Risk

Chapter 4 Return and Risk Chaper 4 Reur ad Risk The objecives of his chaper are o eable you o:! Udersad ad calculae reurs as a measure of ecoomic efficiecy! Udersad he relaioships bewee prese value ad IRR ad YTM! Udersad how obai

More information

Hilbert Transform Relations

Hilbert Transform Relations BULGARIAN ACADEMY OF SCIENCES CYBERNEICS AND INFORMAION ECHNOLOGIES Volume 5, No Sofia 5 Hilber rasform Relaios Each coiuous problem (differeial equaio) has may discree approximaios (differece equaios)

More information

THE IMPACT OF FINANCING POLICY ON THE COMPANY S VALUE

THE IMPACT OF FINANCING POLICY ON THE COMPANY S VALUE THE IMPACT OF FINANCING POLICY ON THE COMPANY S ALUE Pirea Marile Wes Uiversiy of Timişoara, Faculy of Ecoomics ad Busiess Admiisraio Boțoc Claudiu Wes Uiversiy of Timişoara, Faculy of Ecoomics ad Busiess

More information

COLLECTIVE RISK MODEL IN NON-LIFE INSURANCE

COLLECTIVE RISK MODEL IN NON-LIFE INSURANCE Ecoomic Horizos, May - Augus 203, Volume 5, Number 2, 67-75 Faculy of Ecoomics, Uiversiy of Kragujevac UDC: 33 eissn 227-9232 www. ekfak.kg.ac.rs Review paper UDC: 005.334:368.025.6 ; 347.426.6 doi: 0.5937/ekohor30263D

More information

UNIT ROOTS Herman J. Bierens 1 Pennsylvania State University (October 30, 2007)

UNIT ROOTS Herman J. Bierens 1 Pennsylvania State University (October 30, 2007) UNIT ROOTS Herma J. Bieres Pesylvaia Sae Uiversiy (Ocober 30, 2007). Iroducio I his chaper I will explai he wo mos frequely applied ypes of ui roo ess, amely he Augmeed Dickey-Fuller ess [see Fuller (996),

More information

HYPERBOLIC DISCOUNTING IS RATIONAL: VALUING THE FAR FUTURE WITH UNCERTAIN DISCOUNT RATES. J. Doyne Farmer and John Geanakoplos.

HYPERBOLIC DISCOUNTING IS RATIONAL: VALUING THE FAR FUTURE WITH UNCERTAIN DISCOUNT RATES. J. Doyne Farmer and John Geanakoplos. HYPERBOLIC DISCOUNTING IS RATIONAL: VALUING THE FAR FUTURE WITH UNCERTAIN DISCOUNT RATES By J. Doye Farmer ad Joh Geaakoplos Augus 2009 COWLES FOUNDATION DISCUSSION PAPER NO. 1719 COWLES FOUNDATION FOR

More information

A Heavy Traffic Approach to Modeling Large Life Insurance Portfolios

A Heavy Traffic Approach to Modeling Large Life Insurance Portfolios A Heavy Traffic Approach o Modelig Large Life Isurace Porfolios Jose Blache ad Hery Lam Absrac We explore a ew framework o approximae life isurace risk processes i he sceario of pleiful policyholders,

More information

The Norwegian Shareholder Tax Reconsidered

The Norwegian Shareholder Tax Reconsidered The Norwegia Shareholder Tax Recosidered Absrac I a aricle i Ieraioal Tax ad Public Fiace, Peer Birch Sørese (5) gives a i-deph accou of he ew Norwegia Shareholder Tax, which allows he shareholders a deducio

More information

Data Protection and Privacy- Technologies in Focus. Rashmi Chandrashekar, Accenture

Data Protection and Privacy- Technologies in Focus. Rashmi Chandrashekar, Accenture Daa Proeio ad Privay- Tehologies i Fous Rashmi Chadrashekar, Aeure Sesiive Creai Daa Lifeyle o Busiess sesiive daa proeio is o a sigle eve. Adequae proeio o mus be provided appropriaely hroughou Mai he

More information

Granger Causality Analysis in Irregular Time Series

Granger Causality Analysis in Irregular Time Series Grager Causaliy Aalysis i Irregular Time Series Mohammad Taha Bahadori Ya Liu Absrac Learig emporal causal srucures bewee ime series is oe of he key ools for aalyzig ime series daa. I may real-world applicaios,

More information

3. Cost of equity. Cost of Debt. WACC.

3. Cost of equity. Cost of Debt. WACC. Corporae Fiace [09-0345] 3. Cos o equiy. Cos o Deb. WACC. Cash lows Forecass Cash lows or equiyholders ad debors Cash lows or equiyholders Ecoomic Value Value o capial (equiy ad deb) - radiioal approach

More information

ON THE RISK-NEUTRAL VALUATION OF LIFE INSURANCE CONTRACTS WITH NUMERICAL METHODS IN VIEW ABSTRACT KEYWORDS 1. INTRODUCTION

ON THE RISK-NEUTRAL VALUATION OF LIFE INSURANCE CONTRACTS WITH NUMERICAL METHODS IN VIEW ABSTRACT KEYWORDS 1. INTRODUCTION ON THE RISK-NEUTRAL VALUATION OF LIFE INSURANCE CONTRACTS WITH NUMERICAL METHODS IN VIEW BY DANIEL BAUER, DANIELA BERGMANN AND RÜDIGER KIESEL ABSTRACT I rece years, marke-cosise valuaio approaches have

More information

Teaching Bond Valuation: A Differential Approach Demonstrating Duration and Convexity

Teaching Bond Valuation: A Differential Approach Demonstrating Duration and Convexity JOURNAL OF EONOMIS AND FINANE EDUATION olume Number 2 Wier 2008 3 Teachig Bod aluaio: A Differeial Approach Demosraig Duraio ad ovexi TeWah Hah, David Lage ABSTRAT A radiioal bod pricig scheme used i iroducor

More information

Duration and Convexity ( ) 20 = Bond B has a maturity of 5 years and also has a required rate of return of 10%. Its price is $613.

Duration and Convexity ( ) 20 = Bond B has a maturity of 5 years and also has a required rate of return of 10%. Its price is $613. Graduae School of Business Adminisraion Universiy of Virginia UVA-F-38 Duraion and Convexiy he price of a bond is a funcion of he promised paymens and he marke required rae of reurn. Since he promised

More information

Exchange Rates, Risk Premia, and Inflation Indexed Bond Yields. Richard Clarida Columbia University, NBER, and PIMCO. and

Exchange Rates, Risk Premia, and Inflation Indexed Bond Yields. Richard Clarida Columbia University, NBER, and PIMCO. and Exchage Raes, Risk Premia, ad Iflaio Idexed Bod Yields by Richard Clarida Columbia Uiversiy, NBER, ad PIMCO ad Shaowe Luo Columbia Uiversiy Jue 14, 2014 I. Iroducio Drawig o ad exedig Clarida (2012; 2013)

More information

TACTICAL PLANNING OF THE OIL SUPPLY CHAIN: OPTIMIZATION UNDER UNCERTAINTY

TACTICAL PLANNING OF THE OIL SUPPLY CHAIN: OPTIMIZATION UNDER UNCERTAINTY TACTICAL PLANNING OF THE OIL SUPPLY CHAIN: OPTIMIZATION UNDER UNCERTAINTY Gabriela Ribas Idusrial Egieerig Deparme Poifical Caholic Uiversiy of Rio de Jaeiro PUC-Rio, CP38097, 22453-900 Rio de Jaeiro Brazil

More information

SECTION 1.5 : SUMMATION NOTATION + WORK WITH SEQUENCES

SECTION 1.5 : SUMMATION NOTATION + WORK WITH SEQUENCES SECTION 1.5 : SUMMATION NOTATION + WORK WITH SEQUENCES Read Sectio 1.5 (pages 5 9) Overview I Sectio 1.5 we lear to work with summatio otatio ad formulas. We will also itroduce a brief overview of sequeces,

More information

On Motion of Robot End-effector Using The Curvature Theory of Timelike Ruled Surfaces With Timelike Ruling

On Motion of Robot End-effector Using The Curvature Theory of Timelike Ruled Surfaces With Timelike Ruling O Moio of obo Ed-effecor Usig he Curvaure heory of imelike uled Surfaces Wih imelike ulig Cumali Ekici¹, Yasi Ülüürk¹, Musafa Dede¹ B. S. yuh² ¹ Eskişehir Osmagazi Uiversiy Deparme of Mahemaics, 6480-UKEY

More information

Asymptotic Growth of Functions

Asymptotic Growth of Functions CMPS Itroductio to Aalysis of Algorithms Fall 3 Asymptotic Growth of Fuctios We itroduce several types of asymptotic otatio which are used to compare the performace ad efficiecy of algorithms As we ll

More information

PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE

PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE Profi Tes Modelling in Life Assurance Using Spreadshees PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE Erik Alm Peer Millingon 2004 Profi Tes Modelling in Life Assurance Using Spreadshees

More information

Capital Budgeting: a Tax Shields Mirage?

Capital Budgeting: a Tax Shields Mirage? Theoreical ad Applied Ecoomics Volume XVIII (211), No. 3(556), pp. 31-4 Capial Budgeig: a Tax Shields Mirage? Vicor DRAGOTĂ Buchares Academy of Ecoomic Sudies vicor.dragoa@fi.ase.ro Lucia ŢÂŢU Buchares

More information

A simple SSD-efficiency test

A simple SSD-efficiency test A simple SSD-efficiecy es Bogda Grechuk Deparme of Mahemaics, Uiversiy of Leiceser, UK Absrac A liear programmig SSD-efficiecy es capable of ideifyig a domiaig porfolio is proposed. I has T + variables

More information

A Bayesian Based Search and Classification System for Product. Information of Agricultural Logistics Information Technology

A Bayesian Based Search and Classification System for Product. Information of Agricultural Logistics Information Technology A Bayesia Based Searh ad Classifiaio Sysem for Produ Iformaio of Agriulural Logisis Iformaio Tehology Dada Li 1,Daoliag Li 1,3, Yigyi Che 1,3, Li Li 1, Xiagyag Qi 3, Yogu Zheg 1, * 1 Chia Agriulural Uiversiy,

More information

Hedging with Forwards and Futures

Hedging with Forwards and Futures Hedging wih orwards and uures Hedging in mos cases is sraighforward. You plan o buy 10,000 barrels of oil in six monhs and you wish o eliminae he price risk. If you ake he buy-side of a forward/fuures

More information

Predicting Indian Stock Market Using Artificial Neural Network Model. Abstract

Predicting Indian Stock Market Using Artificial Neural Network Model. Abstract Predicig Idia Sock Marke Usig Arificial Neural Nework Model Absrac The sudy has aemped o predic he moveme of sock marke price (S&P CNX Nify) by usig ANN model. Seve years hisorical daa from 1 s Jauary

More information

DBIQ Regulated Utilities Index

DBIQ Regulated Utilities Index db Ide Develome March 2013 DBIQ Ide Guide DBIQ Regulaed Uiliies Ide Summary The DBIQ Regulaed Uiliies Ide ( Uiliies Ide is a rules-based ide aimig o rack he reurs ou of he regulaed uiliies secor i develoed

More information

.04. This means $1000 is multiplied by 1.02 five times, once for each of the remaining sixmonth

.04. This means $1000 is multiplied by 1.02 five times, once for each of the remaining sixmonth Questio 1: What is a ordiary auity? Let s look at a ordiary auity that is certai ad simple. By this, we mea a auity over a fixed term whose paymet period matches the iterest coversio period. Additioally,

More information

The Transport Equation

The Transport Equation The Transpor Equaion Consider a fluid, flowing wih velociy, V, in a hin sraigh ube whose cross secion will be denoed by A. Suppose he fluid conains a conaminan whose concenraion a posiion a ime will be

More information

Unsteady State Molecular Diffusion

Unsteady State Molecular Diffusion Chaper. Differeial Mass Balae Useady Sae Moleular Diffusio Whe he ieral oeraio gradie is o egligible or Bi

More information

0.7 0.6 0.2 0 0 96 96.5 97 97.5 98 98.5 99 99.5 100 100.5 96.5 97 97.5 98 98.5 99 99.5 100 100.5

0.7 0.6 0.2 0 0 96 96.5 97 97.5 98 98.5 99 99.5 100 100.5 96.5 97 97.5 98 98.5 99 99.5 100 100.5 Sectio 13 Kolmogorov-Smirov test. Suppose that we have a i.i.d. sample X 1,..., X with some ukow distributio P ad we would like to test the hypothesis that P is equal to a particular distributio P 0, i.e.

More information

A GLOSSARY OF MAIN TERMS

A GLOSSARY OF MAIN TERMS he aedix o his glossary gives he mai aggregae umber formulae used for cosumer rice (CI) uroses ad also exlais he ierrelaioshis bewee hem. Acquisiios aroach Addiiviy Aggregae Aggregaio Axiomaic, or es aroach

More information

ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS

ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS R. Caballero, E. Cerdá, M. M. Muñoz and L. Rey () Deparmen of Applied Economics (Mahemaics), Universiy of Málaga,

More information

Economics Honors Exam 2008 Solutions Question 5

Economics Honors Exam 2008 Solutions Question 5 Economics Honors Exam 2008 Soluions Quesion 5 (a) (2 poins) Oupu can be decomposed as Y = C + I + G. And we can solve for i by subsiuing in equaions given in he quesion, Y = C + I + G = c 0 + c Y D + I

More information

12. Spur Gear Design and selection. Standard proportions. Forces on spur gear teeth. Forces on spur gear teeth. Specifications for standard gear teeth

12. Spur Gear Design and selection. Standard proportions. Forces on spur gear teeth. Forces on spur gear teeth. Specifications for standard gear teeth . Spur Gear Desig ad selecio Objecives Apply priciples leared i Chaper 11 o acual desig ad selecio of spur gear sysems. Calculae forces o eeh of spur gears, icludig impac forces associaed wih velociy ad

More information

Determining the sample size

Determining the sample size Determiig the sample size Oe of the most commo questios ay statisticia gets asked is How large a sample size do I eed? Researchers are ofte surprised to fid out that the aswer depeds o a umber of factors

More information

General Bounds for Arithmetic Asian Option Prices

General Bounds for Arithmetic Asian Option Prices The Uiversiy of Ediburgh Geeral Bouds for Arihmeic Asia Opio Prices Colombia FX Opio Marke Applicaio MSc Disseraio Sude: Saiago Sozizky s1200811 Supervisor: Dr. Soirios Sabais Augus 16 h 2013 School of

More information

Determinants of Public and Private Investment An Empirical Study of Pakistan

Determinants of Public and Private Investment An Empirical Study of Pakistan eraioal Joural of Busiess ad Social Sciece Vol. 3 No. 4 [Special ssue - February 2012] Deermias of Public ad Privae vesme A Empirical Sudy of Pakisa Rabia Saghir 1 Azra Kha 2 Absrac This paper aalyses

More information

Section 11.3: The Integral Test

Section 11.3: The Integral Test Sectio.3: The Itegral Test Most of the series we have looked at have either diverged or have coverged ad we have bee able to fid what they coverge to. I geeral however, the problem is much more difficult

More information

Testing the Weak Form of Efficient Market Hypothesis: Empirical Evidence from Jordan

Testing the Weak Form of Efficient Market Hypothesis: Empirical Evidence from Jordan Ieraioal Busiess ad Maageme Vol. 9, No., 4, pp. 9- DOI:.968/554 ISSN 9-84X [Pri] ISSN 9-848 [Olie] www.cscaada.e www.cscaada.org Tesig he Wea Form of Efficie Mare Hypohesis: Empirical Evidece from Jorda

More information

Fuzzy Task Assignment Model of Web Services Supplier

Fuzzy Task Assignment Model of Web Services Supplier Advaed Siee ad Tehology eers Vol.78 (Mulrab 2014),.43-48 h://dx.doi.org/10.14257/asl.2014.78.08 Fuzzy Task Assige Model of Web Servies Sulier Su Jia 1,2,Peg Xiu-ya 1, *, Xu Yig 1,3, Wag Pei-lei 2, Ma Na-ji

More information

Bond Valuation I. What is a bond? Cash Flows of A Typical Bond. Bond Valuation. Coupon Rate and Current Yield. Cash Flows of A Typical Bond

Bond Valuation I. What is a bond? Cash Flows of A Typical Bond. Bond Valuation. Coupon Rate and Current Yield. Cash Flows of A Typical Bond What is a bod? Bod Valuatio I Bod is a I.O.U. Bod is a borrowig agreemet Bod issuers borrow moey from bod holders Bod is a fixed-icome security that typically pays periodic coupo paymets, ad a pricipal

More information

TRANSPORT ECONOMICS, POLICY AND POVERTY THEMATIC GROUP

TRANSPORT ECONOMICS, POLICY AND POVERTY THEMATIC GROUP TRANSPORT NOTES TRANSPORT ECONOMICS, POLICY AND POVERTY THEMATIC GROUP THE WORLD BANK, WASHINGTON, DC Traspor Noe No. TRN-6 Jauary 2005 Noes o he Ecoomic Evaluaio of Traspor Projecs I respose o may requess

More information

The Application of Multi Shifts and Break Windows in Employees Scheduling

The Application of Multi Shifts and Break Windows in Employees Scheduling The Applicaion of Muli Shifs and Brea Windows in Employees Scheduling Evy Herowai Indusrial Engineering Deparmen, Universiy of Surabaya, Indonesia Absrac. One mehod for increasing company s performance

More information

Week 3 Conditional probabilities, Bayes formula, WEEK 3 page 1 Expected value of a random variable

Week 3 Conditional probabilities, Bayes formula, WEEK 3 page 1 Expected value of a random variable Week 3 Coditioal probabilities, Bayes formula, WEEK 3 page 1 Expected value of a radom variable We recall our discussio of 5 card poker hads. Example 13 : a) What is the probability of evet A that a 5

More information

Modified Line Search Method for Global Optimization

Modified Line Search Method for Global Optimization Modified Lie Search Method for Global Optimizatio Cria Grosa ad Ajith Abraham Ceter of Excellece for Quatifiable Quality of Service Norwegia Uiversity of Sciece ad Techology Trodheim, Norway {cria, ajith}@q2s.tu.o

More information

Acceleration Lab Teacher s Guide

Acceleration Lab Teacher s Guide Acceleraion Lab Teacher s Guide Objecives:. Use graphs of disance vs. ime and velociy vs. ime o find acceleraion of a oy car.. Observe he relaionship beween he angle of an inclined plane and he acceleraion

More information

On the degrees of irreducible factors of higher order Bernoulli polynomials

On the degrees of irreducible factors of higher order Bernoulli polynomials ACTA ARITHMETICA LXII.4 (1992 On he degrees of irreducible facors of higher order Bernoulli polynomials by Arnold Adelberg (Grinnell, Ia. 1. Inroducion. In his paper, we generalize he curren resuls on

More information

Department of Computer Science, University of Otago

Department of Computer Science, University of Otago Departmet of Computer Sciece, Uiversity of Otago Techical Report OUCS-2006-09 Permutatios Cotaiig May Patters Authors: M.H. Albert Departmet of Computer Sciece, Uiversity of Otago Micah Colema, Rya Fly

More information

where: T = number of years of cash flow in investment's life n = the year in which the cash flow X n i = IRR = the internal rate of return

where: T = number of years of cash flow in investment's life n = the year in which the cash flow X n i = IRR = the internal rate of return EVALUATING ALTERNATIVE CAPITAL INVESTMENT PROGRAMS By Ke D. Duft, Extesio Ecoomist I the March 98 issue of this publicatio we reviewed the procedure by which a capital ivestmet project was assessed. The

More information

IDENTIFICATION OF MARKET POWER IN BILATERAL OLIGOPOLY: THE BRAZILIAN WHOLESALE MARKET OF UHT MILK 1. Abstract

IDENTIFICATION OF MARKET POWER IN BILATERAL OLIGOPOLY: THE BRAZILIAN WHOLESALE MARKET OF UHT MILK 1. Abstract IDENTIFICATION OF MARKET POWER IN BILATERAL OLIGOPOLY: THE BRAZILIAN WHOLESALE MARKET OF UHT MILK 1 Paulo Robero Scalco Marcelo Jose Braga 3 Absrac The aim of his sudy was o es he hypohesis of marke power

More information

Research Article Sign Data Derivative Recovery

Research Article Sign Data Derivative Recovery Iteratioal Scholarly Research Network ISRN Applied Mathematics Volume 0, Article ID 63070, 7 pages doi:0.540/0/63070 Research Article Sig Data Derivative Recovery L. M. Housto, G. A. Glass, ad A. D. Dymikov

More information

Soving Recurrence Relations

Soving Recurrence Relations Sovig Recurrece Relatios Part 1. Homogeeous liear 2d degree relatios with costat coefficiets. Cosider the recurrece relatio ( ) T () + at ( 1) + bt ( 2) = 0 This is called a homogeeous liear 2d degree

More information

Data Analysis and Statistical Behaviors of Stock Market Fluctuations

Data Analysis and Statistical Behaviors of Stock Market Fluctuations 44 JOURNAL OF COMPUTERS, VOL. 3, NO. 0, OCTOBER 2008 Data Aalysis ad Statistical Behaviors of Stock Market Fluctuatios Ju Wag Departmet of Mathematics, Beijig Jiaotog Uiversity, Beijig 00044, Chia Email:

More information

1 Computing the Standard Deviation of Sample Means

1 Computing the Standard Deviation of Sample Means Computig the Stadard Deviatio of Sample Meas Quality cotrol charts are based o sample meas ot o idividual values withi a sample. A sample is a group of items, which are cosidered all together for our aalysis.

More information

Abstract. 1. Introduction. 1.1 Notation. 1.2 Parameters

Abstract. 1. Introduction. 1.1 Notation. 1.2 Parameters 1 Mdels, Predici, ad Esimai f Oubreaks f Ifecius Disease Peer J. Csa James P. Duyak Mjdeh Mhashemi {pjcsa@mire.rg, jduyak@mire.rg, mjdeh@mire.rg} he MIRE Crprai 202 Burlig Rad Bedfrd, MA 01730 1420 Absrac

More information

INTEREST RATE FUTURES AND THEIR OPTIONS: SOME PRICING APPROACHES

INTEREST RATE FUTURES AND THEIR OPTIONS: SOME PRICING APPROACHES INTEREST RATE FUTURES AND THEIR OPTIONS: SOME PRICING APPROACHES OPENGAMMA QUANTITATIVE RESEARCH Absrac. Exchange-raded ineres rae fuures and heir opions are described. The fuure opions include hose paying

More information

Why Did the Demand for Cash Decrease Recently in Korea?

Why Did the Demand for Cash Decrease Recently in Korea? Why Did he Demand for Cash Decrease Recenly in Korea? Byoung Hark Yoo Bank of Korea 26. 5 Absrac We explores why cash demand have decreased recenly in Korea. The raio of cash o consumpion fell o 4.7% in

More information

Output Analysis (2, Chapters 10 &11 Law)

Output Analysis (2, Chapters 10 &11 Law) B. Maddah ENMG 6 Simulatio 05/0/07 Output Aalysis (, Chapters 10 &11 Law) Comparig alterative system cofiguratio Sice the output of a simulatio is radom, the comparig differet systems via simulatio should

More information