5.3.1 Netflix Prize

I-Netflix Prize isebenzisa call evulekile ukuze ukubikezela ukuthi imaphi amabhayisikobho abantu azokuthanda.

Iphrojekthi evulekile kakhulu eyaziwayo evulekile ngumklomelo we-Netflix. I-Netflix inkampani eqashisayo ye-movie, futhi ngonyaka ka-2000 iqalise i-Cinematch, isevisi yokutusa ama-movie kumakhasimende. Isibonelo, i-Cinematch ingase ibone ukuthi uthanda i- Star Wars ne- The Empire Strikes Back bese-ke uncoma ukuthi ubuke ukubuya kwe-Jedi . Ekuqaleni, i-Cinematch isebenze kabi. Kodwa, phakathi neminyaka eminingi, yaqhubeka ithuthukisa ikhono layo lokubikezela ukuthi yiziphi amakhasimende ama-movie azokujabulela. Kodwa ngo-2006, ukuthuthukiswa kweCinematch kwase kuphele. Abacwaningi baseNetflix bazame konke okucabangayo, kodwa ngesikhathi esifanayo, bacabanga ukuthi kunezinye imibono ezingabasiza ukuba bathuthukise uhlelo lwabo. Ngakho-ke, bakhuphuka ngalokho okwakukhona, ngaleso sikhathi, isisombululo esikhulu: ucingo oluvulekile.

Okubalulekile ekuphumeleleni okuphelele komklomelo we-Netflix ukuthi indlela yokuvula ifakwe kanjani, futhi lokhu kuklanywa kunesifundo esibalulekile sokuthi izingcingo ezivulekile zingasetshenziselwa kanjani ucwaningo lwezenhlalakahle. I-Netflix ayifuni nje ukufaka isicelo esingakahleleki semibono, yilokho abantu abaningi abakucabange khona lapho beqala ukucelwa ucingo oluvulekile. Kunalokho, i-Netflix ibe nenkinga ecacile ngenqubo elula yokuhlola: inselele abantu ukusebenzisa isethi yezilinganiso ze-movie eziyizigidi ezingu-100 ukubikezela izilinganiso ezibekelwe izigidi ezingu-3 (izilinganiso abasebenzisi abenzile kodwa ukuthi i-Netflix ayikhiphi). Umuntu wokuqala ukudala i-algorithm eyabikezela ukulinganisa okungu-3 million okukhishiwe okungaphezulu kuka-10% okungcono kunokuba i-Cinematch izowina amadola ayizigidi. Le nqubo ecacile futhi elula yokusebenzisa inqubo yokuhlola-ukuqhathanisa izilinganiso ezibikezelwe nezilinganiso ezigcinwe-kwakusho ukuthi umklomelo we-Netflix uhlelwe ngendlela yokuthi izixazululo zilula ukuhlola kunokukhiqiza; kuphendulele inselele yokuthuthukisa i-Cinematch ibe inkinga efanelekile ikholi evulekile.

Ngo-Okthoba ka-2006, i-Netflix ikhishwe i-dasaset equkethe izilinganiso ze-movie eziyizigidi ezingu-100 ezivela kumakhasimende angaba ngu-500,000 (sizocabangela impendulo yobumfihlo yalokhu kukhishwa kwedatha esahlukweni 6). Idatha ye-Netflix ingacatshangwa njengomthamo omkhulu okungaba amakhasimende abangaba ngu-500,000 ngama-movie angu-20,000. Ngaphakathi kwalesi sikole, kwakukhona izilinganiso ezingaba yizigidi ezingu-100 esikalini kusuka kwinye izinkanyezi ezinhlanu (ithebula 5.2). Inselelo kwakuwukusebenzisa idatha ehlonziwe ku-matrix ukubikezela izilinganiso ezi-3 ezigcinwe ngokubanjwa.

Abacwaningi nabaduni emhlabeni wonke babhekene nenselele, futhi ngo-2008 abantu abangaphezu kuka-30 000 basebenza kulo (Thompson 2008) . Phakathi nomncintiswano, iNetflix yathola izixazululo ezingaphezu kuka-40,000 ezihlongozwayo ezivela emaqenjini angaba ngu-5 000 (Netflix 2009) . Ngokusobala, iNetflix ayikwazanga ukufunda nokuqonda zonke lezi zixazululo ezihlongozwayo. Into yonke yagijima kahle, kodwa, ngoba izixazululo zazilula ukuhlola. I-Netflix ingaba nekhompiyutha kuphela uma iqhathanisa izilinganiso ezibikezelwe ngokulinganisa okugcinwe ngokusebenzisa i-metric ecacisiwe (i-metric ethile ayisebenzisayo yayingumgqa wesigcawu sephutha elibhekiselweyo). Kwakuyikhono lokuhlola masinyane izixazululo ezenza i-Netflix ukwamukela izixazululo ezivela kuwo wonke umuntu, okwakubonakala kubalulekile ngoba imibono emihle yavela kwezinye izindawo ezimangalisayo. Eqinisweni, isisombululo esiphumelele sathunyelwa yiqembu elaqalwa ngabacwaningi abathathu ababengenaso isipiliyoni sezinhlelo zokuncoma ze-movie (Bell, Koren, and Volinsky 2010) .

Esinye isici esihle seNetflix Umklomelo ukuthi senze zonke izixazululo ezihlongozwayo zihlolwe ngendlela efanele. Okusho ukuthi, lapho abantu belayisha izilinganiso zabo ezibikezelwe, akudingeki ukuthi balayishe iziqinisekiso zabo zezemfundo, ubudala, ubuhlanga, ubulili, ukuziphatha ngokocansi noma noma yini ngokwabo. Izilinganiso ezibikezelwe zomprofesa odumile waseStanford zathinteka ngokufanayo nalokho ezisuka kumuntu osemusha ekamelweni lakhe. Ngeshwa, lokhu akulona iqiniso ocwaningweni oluningi lomphakathi. Okungukuthi, ngokucwaninga kwezenhlalakahle, ukuhlolwa kudla isikhathi esiningi futhi kuyingxenye yokuziphendulela. Ngakho-ke, imibono eminingi yokucwaninga ayihlolisiswanga kahle, futhi uma imibono ihlolwa, kunzima ukuthola lokho kuhlolwa kumdali wemibono. Vula amaphrojekthi okushaya ucingo, ngakolunye uhlangothi, abe nokuhlola okulula nokulinganayo ukuze bakwazi ukuthola imibono engaphuthelwa ngenye indlela.

Isibonelo, ngesikhathi esisodwa ngesikhathi somvuzo we-Netflix, othile ophethe igama lesihenqo uSimon Funk uthumele kubhulogi lakhe ikhambi elihlongozwayo elisuselwa ekuqhekekeni kwenani elilodwa, indlela evela e-algebra ehlukile engakaze isetshenziswe ngaphambilini ngabanye abahlanganyeli. Ukuthunyelwa kwebhulogi ye-Funk okwamanje kwakungokwemvelo futhi kungabonakali kahle. Ingabe lokhu okuthunyelwe kubhulogi kuchaza isisombululo esihle noma bekuyikuchitha isikhathi? Ngaphandle kweprojekthi yokuvula ucingo, isixazululo kungenzeka singakaze sithole ukuhlolwa okujulile. Phela, uSimon Funk wayengesiyena profesa ku-MIT; wayengumthuthukisi wesofthiwe, ngaleso sikhathi, owayesebenza emuva kwe-New Zealand (Piatetsky 2007) . Uma ngabe uthumelele lo mqondo ku-Netflix lo mqondo, cishe cishe wawungafundiwe.

Ngenhlanhla, ngoba indlela yokuhlola yayicacile futhi kulula ukuyisebenzisa, ukulinganisa kwakhe okubikezelwe kwahlolwa, futhi kwacaca ngokushesha ukuthi indlela yakhe yayinamandla kakhulu: wayidonsa endaweni yesine emncintiswaneni, umphumela omkhulu owawunikwe amanye amaqembu isebenza izinyanga kule nkinga. Ekugcineni, izingxenye zendlela ayezisebenzisayo zazisetshenziselwa bonke abathintekayo (Bell, Koren, and Volinsky 2010) .

Iqiniso lokuthi uSimon Funk wakhetha ukubhala okuthunyelwe kwebhulogi echaza indlela yakhe, kunokuzama ukuyibeka ngasese, futhi kubonisa ukuthi abahlanganyeli abaningi kwiNetflix Prize abazange batholakale kuphela umklomelo wezigidigidi. Esikhundleni salokho, abahlanganyeli abaningi babonakala bejabulela inselele yengqondo nomphakathi owakhiwe ngale nkinga (Thompson 2008) , imizwa engilindele ukuba abacwaningi abaningi baqonde.

Umklomelo we-Netflix uyisibonelo sokuqala socingo oluvulekile. I-Netflix yenze umbuzo ngomgomo othize (ukubikezela ukulinganisa kwe-movie) futhi yacela izixazululo ezivela kubantu abaningi. I-Netflix yakwazi ukuhlola zonke lezi zixazululo ngoba zazilula ukuhlola kunokudala, futhi ekugcineni i-Netflix ithole ikhambi elihle kakhulu. Okulandelayo, ngizokubonisa ukuthi le ndlela ingasetshenziswa kanjani ku-biology nomthetho, futhi ngaphandle komklomelo we-dollar wezigidi.