5.3.1 Netflix Prize

The Netflix Prize usebenzisa umnxeba evulekileyo ukuqikelela apho bhanya kukuthanda abantu.

Le projekthi eyaziwa kakhulu kakuhle umnxeba evulekileyo ke Prize Netflix. Netflix na inkampani movie ezirentwayo intanethi, yaye ngowe-2000 oko ngokusesikweni Cinematch, inkonzo ukuba licebise bhanya kubathengi. Umzekelo, Cinematch ukuze qaphela ukuba oyithandileyo Star Iimfazwe uBukhosi Ubetha Emva uze ke sicebisa ukuba ukubukela kubuya kwe Jedi. Ekuqaleni, Cinematch wasebenza kakubi. Kodwa ke, ekuhambeni kweminyaka emininzi, Cinematch waqhubeka ukuphucula amandla ayo ukuqikelela ukuba bhanya bangakonwabela abathengi. Ngo-2006, nangona kunjalo, inkqubela Cinematch plateaued. Abaphandi ngexesha Netflix bazama phantse yonke eyayicinga, kodwa kwangaxeshanye, barhanela ukuba kukho ezinye iingcamango ukuze kubanceda baphucule indlela yabo. Ngenxa yoko, bebenyuka yintoni, ngelo xesha, isisombululo IINDLAVINI: ukufowunela evulekileyo.

Ebaluleke kakhulu kwimpumelelo ethubeni Netflix Prize waye sayilwa indlela ifowuni evulekileyo, kwaye oku uyilo izifundo ezibalulekileyo malunga nendlela iminxeba ezivulekileyo zingasetyenziswa kuphando kwezentlalo. Netflix akazange nje phandle isicelo efan ngenxa izimvo, nto leyo abantu abaninzi bacinga xa babeqala ingqalelo umnxeba evulekileyo. Kunoko, Netflix yingxaki ecacileyo kunye novavanyo elula: bona umngeni abantu ukusebenzisa iseti 100 lezigidi amanqaku bhanyabhanya ukuqikelela 3 million ratings zabanjwa-phandle (ratings abasebenzisi wayenze kodwa Netflix akazange ukukhulula). Nabani na inokudala algorithm ukuba akwazi ukuqikelela ratings 3 million kubanjwa-phandle 10% ngaphezu Cinematch uza kuphumelela izigidi 1 zeerandi. Oku ecacileyo kwaye kulula ukuba isicelo novavanyo nqobo-ngokuthelekisa amanqaku icingelwe ukuba kubanjwa-phandle ratings-kwakuthetha ukuba Prize Netflix nto yenziwe ngendlela yokuba izisombululo kulula ukukhangela kunokuba ukujikelezisa; kulwenze mngeni yokuphucula Cinematch ibe yingxaki esifanelekileyo kukho ifowuni evulekileyo.

Ngo-Oktobha ka-2006, Netflix ikhuphe dataset equlathe 100 lezigidi ratings kamabonakude malunga 500,000 (siza kuqwalasela iziphumo zabucala sokukhululwa data kwiSahluko 6). Idatha Netflix ibe nengcinga njenge matrix enkulu ukuba malunga nama-500,000 abathengi ziimovie 20,000. Ngaphakathi kule isizalo, kwakukho malunga 100 lezigidi amanqaku ngokomlinganiselo kweenkwenkwezi-1 ukuya kwisi-5 (Uludwe 5.2). Olona celomngeni ukusebenzisa data alibona isizalo ukuqikelela ratings 3 million kubanjwa-phandle.

Itheyibhile 5.2: Umzobo data evela Netflix Prize. Netflix yakhululwa ukureyithwa 100 lezigidi (iinkwenkwezi inkwenkwezi 1 ukuya 5) ezinikwa-500,000 abathengi iimovie 20,000. Injongo Netflix Prize wayeza ukusebenzisa ezi amanqaku ukuqikelela ratings kubanjwa kwe-3 sezigidi eshukumayo, iboniswe '? ". amanqaku eziqikelelweyo ezingeniswe nxaxheba kwi Netflix Prize zaye kuthelekiswa ratings kubanjwa-phandle. Ndiya kuxoxa imiba esesikweni ezingqonge sokukhululwa data kwiSahluko 6.
bhanya 1 bhanya 2 bhanya 3 . . . bhanya 20,000
abathengi 1 2 5 . ?
abathengi 2 2 ? . 3
abathengi 3 ? 2 .
. . . . . . . .
abathengi 500,000 ? 2 . 1

Abaphandi kunye imigewu ehlabathini lonke batsaleleke umngeni, yaye ngo-2008 abantu abangaphezu kwama-30,000 ababesebenza kuyo (Thompson 2008) . Kule khosi lokhuphiswano, Netflix wafumana ngaphezulu kwama-40,000 izisombululo ecetywayo amaqela ngaphezulu kwama-5,000 (Netflix 2009) . Ngokucacileyo, Netflix wayengakwazi ukufunda nokuqonda zonke ezi sisombululo siphakanyiswayo. yonke le nto wabaleka kakuhle, Noko ke, ngenxa yokuba kwizisombululo lula ukutshekisha. Netflix nako nje ne khompyutha ukuthelekisa amanqaku icingelwe ukuba amanqaku okuhlela ezigcinwe-ophuma metric (obala ethile abayisebenzisileyo yaba ingcambu esikweri yesiphoso kuthetha-leskwere) pre-ekhankanyiweyo. Ukuba oku ukukwazi ukuhlola ngokukhawuleza izisombululo eyanceda Netflix ukwamkela izisombululo ukusuka wonke, nto yeza ukuba kubalulekile kuba iingcamango elungileyo ezinye iindawo ezimangalisayo. Enyanisweni, isicombululo eliphumelelayo yangeniswa liqela uqalwa abaphandi ezintathu kungekho amava okwakha iinkqubo movie isindululo phambi (Bell, Koren, and Volinsky 2010) .

Enye into entle Netflix Prize kukuba Anceda wonke umntu ehlabathini ukuba isisombululo yabo avavanywe ngokufanelekileyo. Xa abantu kufakwa amanqaku zabo kwangaphambili, akazange akuyomfuneko ukuba ifayile klk yabo ezifundweni, yabo ubudala, uhlanga, isini, ubuni, okanye nantoni na malunga ngokwabo. Ngoko ke, amanqaku ezicingelwayo unjingalwazi odumileyo evela Stanford baphathwa ncam naleyo evela elivisayo egumbini lakhe lokulala. Ngelishwa, oku akuyonyaniso ezininzi zophando lwentlalo. Oko kukuthi, kuba inkoliso uphando lwentlalo, uvavanyo kungatya ixesha elide ngokwenxalenye ayinyani. Ngoko ke, inkoliso iingcamango uphando soze kuvavanywa kakhulu, kwaye xa kuvavanywa iingcamango, kunzima ukuba angozohlula ezo iimvavanyo ukusuka umdali izimvo. Ngenxa izisombululo kulula ukukhangela, iminxeba ezivulekileyo kuvumela abaphandi ukuba ukufikelela zonke izisombululo ezinokuba ezimangalisayo ukuba uza kuwa ngenxa iintanda xa kuqwalaselwa babhenela oonjingalwazi abaziwayo kuphela.

Ngokomzekelo, ngaxa lithile ebudeni umntu Netflix Prize kunye negama leskrini Simon Funk ngeposi kwi blog isisombululo sakhe ecetywayo ngokusekelwe isinye ixabiso yokubola, indlela ukusuka algebra yomgama ukuba ayisetyenziswanga ngaphambili ngamanye nxaxheba. ngeposi blog Funk yaba ngaxeshanye yobugcisa Isimanga sesikweni. Ngaba esi sithuba blog echaza isisombululo elungileyo okanye kube yinkcitha xesha? Ngaphandle iprojekthi umnxeba evulekileyo, isicombululo kuze ufumene uvavanyo ezinzulu. Emva kokuba yonke Simon Funk wayengekho unjingalwazi Cal Tech okanye MIT; ukuba umakhi isoftwe owathi, ngelo xesha, ukuba Backpacking ngeenxa New Zealand (Piatetsky 2007) . Ukuba ke imeyile kule ingcamango njengenjineli Netflix, oko phantse ngokuqinisekileyo kwakungayi ziye zithathwe ngondiliseko.

Ngethamsanqa, ngenxa yokuba inkqubo yohlolo zazicacile kwaye kulula ukuba isicelo, zavavanywa iireyithingi kwangaphambili, yaye ngoko nangoko kwacaca ukuba indlela yakhe enamandla kakhulu: yena ndle kwindawo yesine kukhuphiswano, isiphumo engathethekiyo enikwe ukuba amanye amaqela bekusele ukusebenza iinyanga ngxaki. Ekugqibeleni, iinxalenye indlela Simon Funk ngayo kwasetyenziswa phantse zonke okhuphisana ezinzulu (Bell, Koren, and Volinsky 2010) .

Inyaniso ukuba uSimon Funk wakhetha ukubhala isithuba blog echaza indlela yakhe, kunokuba izama ukugcina kuyimfihlelo, naye ubonisa ukuba abathathi-nxaxheba abaninzi Netflix Prize azizange kuphela beqhutywa ibhaso sigidi se dollar. Kunoko, nxaxheba abaninzi kwabonakala onwabele mngeni ngokwasengqondweni uluntu ukuba kuphuhliswa ngeenxa ingxaki (Thompson 2008) , iimvakalelo ukuba ndilindele ukuba abaphandi abaninzi baqonde.

The Netflix Prize ngumzekelo classic ngomnxeba evulekileyo. Netflix wabuza umbuzo usukelo ethile (ukuqikelela amanqaku movie) kwaye igaleleo izisombululo kubantu abaninzi. Netflix wakwazi ukuhlola zonke ezi izisombululo ngokuba lula ukuqinisekisa kunokuba ukudala, yaye ekugqibeleni Netflix wakhetha simse. Okulandelayo, ndiza kukubonisa indlela le ndlela inye ingasetyenziswa zebhayoloji kunye nomthetho.