EmploymentRecommendationSystem:AReview

Page 1

I nt er nat i onal Jour nal ofSci enceandEngi neer i ngAppl i cat i ons Vol ume7–No. 5, 6467, May2018, I SSN: 2319-7560

Empl oy mentRecommendat i onSy st em:ARev i ew RoshanG.Bel sar e Depar t mentofComput er Engi neer i ng

Dr .V.M.Deshmukh Depar t mentofComput er Engi neer i ng PRMI TR, Badner a Mahar asht r a, I ndi a v mdeshmukh@mi t r a. ac. i n

PRMI TR, Badner a Mahar asht r a, I ndi a r oshanbel sar e@gmai l . com

Abst r act : Enor mousamount sofj obsar epost edont hej obwebsi t esondai l ybasi sandl ar genumber sofnewr esumesar e al soaddedt oj obwebsi t esdai l y .I nsuchscenar i oi t ’ sav er yt oughj obt osuggestmat chi ngj obst ot hej obappl i cant s.A r ecommendat i onsy st em cansol v et hi spr obl em t ot hegr eatex t ent .Ar ecommendat i onsy st em hasal r eadybeenpr ov ed t obev er yef f ect i v ei nt hear eaofOnl i neshoppi ngwebsi t esandMov i er ecommendat i on.Gi v enauser ,t hegoalofan empl oy mentr ecommendat i onsy st em i st opr edi ctt hosej obposi t i onst hatar el i kel yt ober el ev antt ot heuser .An Empl oy mentr ecommendat i onsy st em woul dsuggestmat chi ngj obst ot heuser susi ngmat chi ng,col l abor at i v ef i l t er i ng andcont entbasedr ecommendat i onbasedonr anki ng. Key wor ds: r ecommendat i onsy st em, col l abor at i v ef i l t er i ng, cont entbasedf i l t er i ng, mat chi ng

1.

I NTRODUCTI ON

The r ecommender sy st em t echnol ogy pl ay s an i mpor t antr ol ei nv ar i ousecommer ceappl i cat i onsby hel pi ngi ndi v i dual st of i ndr i ghti t emsi nal ar geopt i on space,whi ch mat ch t hei ri nt er est s.Recommender sy st emsar esof t war et ool sandt echni quespr ov i di ng suggest i onsf ori t emst obeofi nt er estt oausersuchas v i deos,songs,ornewsar t i cl es.Dr i v enbyt hi ssuccess, mor eappl i cat i ondomai nshav eadopt edr ecommender sy st ems t o r educe t he i nf or mat i on ov er l oad by gener at i ng per sonal i zed suggest i ons. Al so f or r ecr ui t mentscenar i os,i nwhi chappl i cant ssear chf or sui t abl ej obof f er s,r ecommendersy st emsar eausef ul t oolf orj obcandi dat es,r ecr ui t er s,aswel laspl at f or ms t hatconnectbot h[ 12] .Themai nst r eam appr oachest o r ecommender sy st ems ar e cl assi f i ed i nt o f our cat egor i es: Col l abor at i v e Fi l t er i ng ( CF) ,Cont ent Based Fi l t er i ng ( CBF) ,knowl edgebased and hy br i d appr oaches [ 1] . Besi des, ut i l i t y basedanddemogr aphi cappr oachesal so exi st .Themai nadv ant ageofCFappr oachesi st hatt hey canf i ndt hepat t er nsamonguserr at i ngsdat aandwor k wel l f or compl ex obj ect s. [ 2] The pr obl em of r ecommendi ngj obst ouser si sf undament al l ydi f f er ent f r om t r adi t i onal r ecommendat i onsy st em pr obl emssuch asr ecommendi ngbooks,pr oduct s,ormov i est ouser s. Whi l eal loft heabov ehav ea common obj ect i v et o maxi mi zet heengagementr at eoft heuser s,onekey di f f er encei st hataj obpost i ngi st y pi cal l ymeantt ohi r e oneoraf ew empl oy eesonl y ,wher east hesamebook, pr oduct ,ormov i ecoul dbepot ent i al l yr ecommendedt o hundr edsoft housandsofuser sf orconsumpt i on. [ 13] I dealj ob r ecommendat i on sy st em woul d need t o achi ev et hr eegoal ssi mul t aneousl y :( 1)Recommendt he mostr el ev antj obst ouser s.( 2)Ensur et hateachj ob post i ngr ecei v essuf f i ci entnumberofappl i cat i onsf r om qual i f i edcandi dat es.

2.

acr oss many web appl i cat i ons, e. g. , mov i e r ecommendat i ons [ 14] , ecommer ce i t em r ecommendat i ons[ 15] ,j obr ecommendat i ons[ 16]and so f or t h,wher e aut hor s mai nl y concent r at e on t he r el ev ance r et r i ev al and r anki ng aspect s of t he r ecommendat i onsy st em.Ther ei si nsi ght f ulr esear ch and model i ng of t he hi r i ng pr ocesses wi t hi nj ob mar ket pl aces.Suchr esear chi ncl udeswor kr el at edt o est i mat i on ofempl oy ee r eput at i on f oropt i malhi r i ng deci si ons[ 17] ,aswel laswor kr el at edt or anki ngand r el ev ance aspect s ofj ob mat chi ng i nl abormar ket pl aces[ 18,19,20] .Ther ehasbeenwor kr el at edt ot he t heor yofopt i malhi r i ngpr ocess,e. g. ,ont hepr obl em of f i ndi ngt her i ghthi r ef oraj ob( t hehi r i ngpr obl em) ,as wel lason t he cl assi calsecr et ar ypr obl em,wher ea gr owi ngcompanycont i nuousl yi nt er v i ewsanddeci des whet hert o hi r e appl i cant s[ 21,22] . Aut hor s of[ 23] i nv est i gat edj obmar ket pl aceasat wosi dedmat chi ng mar ketusi ng l ocal l y st abl e mat chi ng al gor i t hms f or sol v i ngt hepr obl em off i ndi nganew j obusi ngsoci al cont act s.

3.

Ty pesofRecommenderSy st ems:

Di st i nct i on off ourbasi c al gor i t hm t y pes has been pr oposedi nRS[ 6] : 3. 1 Col l abor at i v e Fi l t er i ng ( CF) Recommender s: Theyut i l i zesoci alknowl edge( t y pi cal l yr at i ngsofi t ems byacommuni t yofuser s)t ogener at er ecommendat i ons. Anew useri smat chedagai nstadat abaset odi scov er nei ghbor s,i . e.ot heruser swho,hi st or i cal l y ,hadsi mi l ar i nt er est s wi t h hi m. Thus,t he i t ems t hat hi s/ her nei ghbor sl i kedar er ecommendedt ot heuserbecause he/ shewi l lpr obabl yl i ket hem t oo[ 7] .Fi g.1i l l ust r at es

Rel at edWor k

JobRecommendat i onwor kr esi desi nt hedomai nof onl i ner ecommendersy st ems, whi char ewi del yadopt ed

www. i j sea. com

64


I nt er nat i onal Jour nal ofSci enceandEngi neer i ngAppl i cat i ons Vol ume7–No. 5, 6467, May2018, I SSN: 2319-7560 t heCFRecommendat i onconcept .[ 11]

knowl edget ogener at er ecommendat i ons. 3. 4

Hy br i dRecommendersy st ems:

Hy br i d Recommendersy st ems combi net woor mor et echni quest ogai nbet t err esul t swi t hf ewer dr awbacks.

4.

Empl oy mentr ecommendat i on sy st em chal l enges

Fi g.1.User1sel ect edA, BandXi t ems, User2sel ect ed AandB, t heCFsy st em wi l l suggestXi t em t oUser2 Col l abor at i v ef i l t er i ng can be cat egor i zed i nt o basi c t y pesnamel yI t em basedcol l abor at i v ef i l t er i nganduser based col l abor at i v e f i l t er i ng. I n t he i t em based col l abor at i v ef i l t er i ng si mi l ari t ems ar ef i nd outt o r ecommend i tt o user s and i n Cont ent based r ecommendat i onuser spastact i v i t i esar eanal y zedt o suggestnewr ecommendat i ons. 3. 2 Cont ent Based ( CB) Recommender s:They ut i l i ze i t em f eat ur es t or ecommend i t ems si mi l art o t hoseauserhasl i kedi nt hepast .ACBsy st em anal y zes asetofchar act er i st i csofi t emst hatar er at edbyauser andbui l dt hepr of i l eoft heuseri nt er est sbasedont he f eat ur esoft hei t emst hatar er at ed byher[ 7] .The r ecommendat i onpr ocessmat chesupt heat t r i but esof t heuserpr of i l eagai nstt hesetofpr oper t i esofacont ent i t em [ 8] , [ 9] , [ 10] .

The j ob mat chi ng pr ocess nor mal l y t akes i nt o consi der at i onoft hedat aav ai l abl ei nt her esumeand mat ch agai nstt he dat al i st ed i nt he l i stofopen v acanci es.Oneoft hemostchal l engi ngt asksoft hi s t y peofj obmat chi ngi st hatt her ear eusual l yt oomany dat at omat chagai nst .Fur t her mor e,t hesedat ausual l y submi t t edi nf r eef or m,aseachi ndi v i dualhast hei rown pr ef er encet opr epar et hedat a.Forexampl e,Per sonA cl ai medt hathehasat ot alof8y ear s‘ exper i encei n Or acl epr oduct .ADat abaseAdmi ni st r at orwi l li nt er pr et t hatt hi sper sonhas8y ear sofexper i encei nOr acl e Dat abase.APl at f or m Leaderwi l lmakeassumpt i ont hat i t ' s8y ear sofexper i encei nOr acl eCommer cePl at f or m, whi l eaLeadPr ogr ammerwi l lt hi nkt hati sa8y ear sof exper i ence i n Jav a pr ogr ammi ng. Ther ef or e, an i nt el l i gentj obmat chi ngengi nei sr equi r edt oov er come t hi si ssue. Recommendi ngaj obi sdi f f er entt hanr ecommendi nga pr oductormov i es as i ti nv ol v es l ar ge numberof par amet er sandf i l t er i ngbasedondi f f er entcr i t er i a. Asamej obcannotber ecommendedt oal lt hepeopl eal l ov ert hewor l dasdemogr aphi car eaal soneedst obe consi der edf orr ecommendat i onofapar t i cul arj obt o par t i cul aruser .

5.

Empl oy ment r ecommendat i on usi ngCol l abor at i v eFi l t er i ng

The t r adi t i onalI t embased CF pr ocesses as f ol l ow: Fi r st , f oreachj obwhi chcur r entuser( user i )appl i edi nt he past( wer egar duser appl i edj obsasuser l i kedj obs) , f i ndoutot heruser swhoappl i ed t hi sj ob( user j )(we r egar dt heseuser sas coappl i eduser s) ,andt henf i nd outot her j obst hese coappl i eduser sal soappl i ed, exceptf ort he cur r entj ob( user l i kedj obs) ,usest hese j obs as candi dat e set . The pr ocedur ei s pr esent ed bel ow:

Fi g. 2.TheCB r ecommenderut i l i zesi t em f eat ur est o r ecommendi t emssi mi l art ot hoseauserhasl i kedi n t hepast Col l abor at i v ef i l t er i ngcanwor k 3. 3 Knowl edgebased ( KB) Recommender s: Knowl edgebased ( KB) Recommender s use domai n

www. i j sea. com

f oreachj obiuser ia ppl i ed { f oreachcoappl i eduser jwh oappl i edj obi { f i ndoutj obs( j obj)t hatuser ja ppl i ed; addt hesej obst ocandi dat eset ; } del et ej obif r om candi dat eset ; } Second, f orev er yj ob I t emj i n t he candi dat e set { I t em1…I t emp} ,comput et hepr edi ctpr ef er encegr ade f ori tusi ng The Jaccar d si mi l ar i t yi ndex.Jaccar d

65


I nt er nat i onal Jour nal ofSci enceandEngi neer i ngAppl i cat i ons Vol ume7–No. 5, 6467, May2018, I SSN: 2319-7560 si mi l ar i t yi scal cul at edusi ngf or mul a: Numberofusercommonf orj obiandj obj( i nt er sect i on) di v i dedbynumberofuser sei t herf or j obi or j obj ( uni on) Atl ast ,sor tal lt hegr adesandchooset opNj obsas t her esul tset .Thepr ocedur ei spr esent edbel ow:

f oreachi t emji ncandi dat eset { comput epr ef( Ui, I t emj) ; } sor tt hesepr ef( Ui, I t em) ;

6.

Empl oy ment r ecommendat i on usi ngCont entbasedFi l t er i ng

bet ween( hi st or i cal l y )appl i edj obsandsi mi l arf eat ur es among new j ob oppor t uni t i es f or consi der at i on. Gener al l yspeaki ng, t hegoalofcont entbasedf i l t er i ngi s t o def i ne r ecommendat i ons based upon f eat ur e si mi l ar i t i es bet ween t he i t ems bei ng consi der ed and i t emswhi chauserhaspr ev i ousl yr at edasi nt er est i ng. f ort het ar getuser i t em r at i ngf ( u,i),cont ent based f i l t er i ng woul d pr edi ctt he opt i malr ecommendat i on basedont heut i l i t yf unct i onsoff ( u,Ih)whi chi st he hi st or i calr at i ngi nf or mat i onofuser uoni t ems(Ih) si mi l arwi t h i.[ 23]Gi v ent hei ror i gi nsoutoft hef i el ds ofi nf or mat i onr et r i ev alandi nf or mat i onf i l t er i ng,most cont ent basedf i l t er i ngsy st emsar eappl i edt oi t emst hat ar er i chi nt ext ual i nf or mat i on.

Oneoft hemostpopul arr ecommenderappr oachesi s cont ent -based f i l t er i ng,whi ch expl oi t st he r el at i ons

7.

REFERENCES

[ 1]K.Wei , J.Huang, andS.Fu.Asur v eyofecommer ce r ecommender sy st ems. I n 2007 I nt er nat i onal Conf er ence on Ser v i ce Sy st ems and Ser v i ce Management , pages1{ 5, June2007. [ 2]Chenr uiZhang,XueqiCheng AnEnsembl eMet hod f orJobRecommenderSy st ems.RecSy sChal l enge’ 16, Sept ember152016, Bost on, MA, USA 2016ACM [ 3]N.D.Al mal i s,G.A.Tsi hr i nt zi sandN.Kar agi anni s," A cont entbasedappr oachf orr ecommendi ngper sonnel f orj ob posi t i ons, "I I SA 2014,The 5t hI nt er nat i onal Conf er enceonI nf or mat i on,I nt el l i gence,Sy st emsand Appl i cat i ons, Chani a, 2014, pp.4549. [ 4]M.Bal abanov i c, andY.Shoham, “ Fab:Cont ent based, Col l abor at i v eRecommendat i on.Communi cat i onsoft he ACM, ”v ol .40, no.3, pp.66-72, 1997. [ 5]Sov r en Gr oup,“ Ov er v i ew oft he Sov r en Semant i c Mat chi ng Engi ne and Compar i son t o Tr adi t i onal Key wor dSear chEngi nes, ”Sov r enGr oupI nc, 2006. [ 6] M.Ramezani ,L.Ber gman,R.Thompson,RBur ke, and B. Mobasher , “ Sel ect i ng and Appl y i ng Recommendat i on Technol ogy , ” I n pr oceedi ngs of I nt er nat i onal Wor kshop on Recommendat i on and Col l abor at i oni nConj uct i onwi t hI nt er nat i onalACM on I nt el l i genceUserI nt er f ace, 2008.

[ 10]RJ.Mooney and L.Roy ,“ Cont ent Based Book Recommendi ngUsi ngLear ni ngf orTextCat egor i zat i on, ” i nPr oceedi ngsofDL‟ 00:Pr oceedi ngsoft heFi f t hACM Conf er enceonDi gi t alLi br ar i es,NewYor k,NY,ACM pp. 195204, 2000. [ 11]N.D.Al mal i s,G.A.Tsi hr i nt zi sandN.Kar agi anni s, " A cont ent based appr oach f or r ecommendi ng per sonnel f or j ob posi t i ons, "I I SA 2014,The 5t h I nt er nat i onalConf er ence on I nf or mat i on,I nt el l i gence, Sy st emsandAppl i cat i ons, Chani a, 2014, pp.4549. [ 12]Toon De Pessemi er ,Kr i s Vanhecke,and Luc Mar t ens.2016.Ascal abl e,hi ghper f or manceAl gor i t hm f orhy br i dj obr ecommendat i ons.I n Pr oceedi ngsoft he

RecommenderSy st ems Chal l enge ( RecSy s Chal l enge ' 16) .ACM,New Yor k,NY,USA,Ar t i cl e5,4pages.DOI : ht t ps: / / doi . or g/ 10. 1145/ 2987538. 2987539 [ 13]Fedor Bor i sy uk,Li ang Zhang,and Kr i shnar am Kent hapadi .2017.Li JAR:ASy st em f orJobAppl i cat i on Redi st r i but i ont owar dsEf f i ci entCar eerMar ket pl ace.I n Pr oceedi ngsofKDD’ 17, Hal i f ax , NS, Canada, August1317,

2017,

10

pages.

ht t ps: / / doi . or g/ 10. 1145/ 3097983. 3098028 [ 14]Car l osA.GomezUr i beandNei lHunt .2015.The

[ 7]Badul Sar war ,G.Kar y pi s,J.Konst an,andJ.Ri edl , “ I t emBased Col l abor at i v e Fi l t er i ng Recommendat i on Al gor i t hms, ” Pr oceedi ngs of t he 10t hI nt er nat i onal Conf er enceofWor l dWi deWeb, pp.285295, 2001.

Net f l i x RecommenderSy st em:Al gor i t hms,Busi ness

[ 8] G.Li nden,B.Smi t h,and J.Yor k,“ Amazon. com Recommendat i ons:I t emt oI t em Col l abor at i v eFi l t er i ng, ” I EEEI nt er netComput i ng, v ol .7, no.1, pp.76–80, 2003

[ 15]Gr egLi nden,Br entSmi t h,andJer emyYor k.2003.

[ 9]D.Ml adeni c,“ Text l ear ni ng and Rel at ed I nt el l i gent Agent s: ASur v ey , ”I EEEI nt el l i gentSy st ems, v ol .14, no.4, pp.44–54, 1999.

www. i j sea. com

Val ue,andI nnov at i on.ACM Tr ans.Manage.I nf .Sy st . ( 2015) .ht t ps: / / doi . or g/ 10. 1145/ 2843948

Amazon. Com

Recommendat i ons: I t emt oI t em

Col l abor at i v e Fi l t er i ng.I EEE I nt er netComput i ng 7,1 ( 2003) ,

76–80.

ht t ps: / / doi . or g/ 10. 1109/ MI C. 2003. 1167344

66


I nt er nat i onal Jour nal ofSci enceandEngi neer i ngAppl i cat i ons Vol ume7–No. 5, 6467, May2018, I SSN: 2319-7560 [ 16]FedorBor i sy uk, Kr i shnar am Kent hapadi , Dav i dSt ei n, andBoZhao.2016.CaSMoS: AFr amewor kf orLear ni ng Candi dat eSel ect i onModel sov erSt r uct ur edQuer i esand Document s.

I n

KDD.

ht t ps: / / doi . or g/ 10. 1145/ 2939672. 2939718 [ 17]Mar i aDal t ay anni ,LucadeAl f ar o,andPanagi ot i s Papadi mi t r i ou. 2015. Wor ker Rank: Usi ng Empl oy er I mpl i ci tJudgement st oI nf erWor kerReput at i on.I n WSDM.ht t ps: / / doi . or g/ 10. 1145/ 2684822. 2685286 [ 18]Vi etHaThuc, YeXu, Sat y aPr adeepKandur i , Xi anr en Wu,Vi j ayDi al ani ,YanYan,Abhi shekGupt a,andShakt i Si nha. 2016. Sear ch by I deal Candi dat es: Next Gener at i on ofTal entSear ch atLi nkedI n.I n WWW. ht t ps: / / doi . or g/ 10. 1145/ 2872518. 2890549 [ 19]Mar i os Kokkodi s,Panagi ot i s Papadi mi t r i ou,and Panagi ot i sG.I pei r ot i s.2015.Hi r i ngBehav i orModel sf or Onl i ne

Labor

Mar ket s.

I n

WSDM.

ht t ps: / / doi . or g/ 10. 1145/ 2684822. 2685299 [ 20]Ji aLi ,Dhr uvAr y a,Vi etHaThuc,andShakt iSi nha. 2016.How t oGetThem aDr eam Job? :Ent i t y Awar e Feat ur esf orPer sonal i zedJobSear chRanki ng.I nKDD. ht t ps: / / doi . or g/ 10. 1145/ 2939672. 2939721 [ 21]Andr eiZ.Br oder , Adam Ki r sch, Rav iKumar , Mi chael Mi t zenmacher ,El iUpf al ,andSer geiVassi l v i t ski i .2008. TheHi r i ngPr obl em andLakeWobegonSt r at egi es.I n SODA.ht t ps: / / doi . or g/ 10. 1137/ 07070629X [ 22]Rav iKumar ,Si l v i oLat t anzi ,Ser geiVassi l v i t ski i ,and Andr eaVat t ani .2011.Hi r i ngaSecr et ar yf r om aPoset .I n EC.ht t ps: / / doi . or g/ 10. 1145/ 1993574. 1993582 [ 23]Est eban Ar caut e and Ser geiVassi l v i t ski i .2009. Soci alNet wor ks and St abl e Mat chi ngs i nt he Job Mar ket .I n WI NE. ht t ps: / / doi . or g/ 10. 1007/ 9783642108419_ 21 [ 24]Shuo Yang a,Mohammed Kor ay em b ,Khal i f eh Al Jadda,Tr eyGr ai nger ,Sr i r aam Nat ar aj anCombi ni ng 2017:cont ent basedandcol l abor at i v ef i l t er i ngf orj ob r ecommendat i on sy st em:A cost sensi t i v e St at i st i cal Rel at i onal

Lear ni ng

appr oach

ht t p: / / dx. doi . or g/ 10. 1016/ j . knosy s. 2017. 08. 017

www. i j sea. com

67


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.