5.3.1 Netflix nqi zog

Lub Netflix nqi zog siv qhib kev hu mus rau kwv yees uas movies neeg yuav nyiam.

Qhov tseem paub zoo qhib hu project yog Netflix nqi zog. Netflix yog ib tug nyob hauv internet movie koj qev los lub tuam txhab, thiab nyob rau hauv 2000 nws launched Cinematch, ib qhov kev pab rau kom cov tsos rau cov neeg muas zaub. Piv txwv li, Cinematch yuav pom hais tias koj nyiam Star Wars thiab faj tim teb chaws tsoo Back thiab ces xav kom koj saib Rov qab ntawm lub Jedi. Chiv, Cinematch ua hauj lwm tsis zoo. Tab sis, nyob rau thaum kawg ntawm ntau xyoo, Cinematch txuas ntxiv mus kom nws muaj peev xwm mus twv seb yog dab tsi movies cov neeg muas zaub yuav txaus siab rau. By 2006, Txawm li cas los, kev kawm rau Cinematch plateaued. Cov soj ntsuam ntawm Netflix tau sim zoo nkauj npaum li cas txhua yam lawv yuav xav seb, tab sis nyob rau tib lub sij hawm, lawv xav tias hais tias muaj lwm yam tswv yim uas yuav pab tau lawv kom lawv system. Yog li, lawv tuaj mus nrog dab tsi yog, thaum lub sij hawm, ib tug radical tshuaj: ib qho qhib hu.

Tseem ceeb heev rau lub eventual kev vam meej ntawm lub Netflix nqi zog yog yuav ua li cas qhib hu tau tsim, thiab qhov no tsim muaj ib qho tseem ceeb qhia rau yuav ua li cas qhib hu yuav siv tau rau kev sib raug zoo kev tshawb fawb. Netflix tsis cia li muab tso tawm ib qho kev unstructured thov rau tswv yim, uas yog dab tsi muaj coob tus neeg xav txog tej yam thaum lawv thawj xav txog ib qho qhib hu. Es, Netflix nyiag ib tug ntshiab teeb meem nrog ib tug yooj yim tshuaj ntsuam xyuas no: lawv sib tw neeg siv ib tug txheej ntawm 100 lab movie kev twv seb 3 lab nyob tawm kev (kev uas cov neeg siv tau ua tab sis hais tias Netflix tsis tso). Neeg twg uas yuav tsim tau ib lub algorithm uas yuav twv seb lub 3 lab nyob tawm kev 10% zoo dua Cinematch yuav yeej 1 lab nyiaj. Qhov no meej thiab yooj yim mus thov kev ntsuam xyuas cov txheej xwm-piv kwv yees kev mus tuav-tawm kev-meant hais tias cov Netflix nqi zog twb framed nyob rau hauv xws li ib tug txoj kev uas daws teeb meem yog yooj yim mus saib dua kom muaj; nws muab cov kev sib tw ntawm kev txhim kho Cinematch rau hauv ib qho teeb meem haum rau ib tug qhib hu.

Nyob rau hauv Lub kaum hli ntuj ntawm 2006, Netflix tso ib tug dataset muaj 100 lab movie kev los txog txog 500,000 cov neeg muas zaub (peb yuav xav txog tus kheej uas siv cov ntaub ntawv no tso tawm nyob rau hauv Tshooj 6). Lub Netflix cov ntaub ntawv yuav tsum conceptualized raws li ib tug lossis loj matrix uas yog kwv yees li 500,000 cov neeg muas zaub los ntawm 20,000 movies. Tsis pub dhau no matrix, muaj txog 100 lab kev rau ib tug scale ntawm 1 mus rau 5 hnub qub (Rooj 5.2). Cov kev sib tw yog siv cov cai ntaub ntawv nyob rau hauv lub matrix twv seb lub 3 lab nyob tawm kev.

Rooj 5.2: Schematic ntawm cov ntaub ntawv los ntawm lub Netflix nqi zog. Netflix tso txog 100 lab kev (1 lub hnub qub rau 5 hnub qub) yog muab los ntawm 500,000 cov neeg muas zaub rau 20,000 movies. Lub hom phiaj ntawm lub Netflix nqi zog yog siv cov kev mus twv seb tus tuav-tawm kev ntawm 3 lab movies, qhia raws li "?". Predicted kev xa los ntawm koom nyob rau hauv lub Netflix nqi zog tau muab piv rau cov nyob tawm kev. Kuv yuav tham txog qhov kev zoo thiab cov teeb meem uas lwm tus cov ntaub ntawv no tso tawm nyob rau hauv Tshooj 6.
movie 1 movie 2 movie 3 . . . movie 20,000
neeg yuav khoom 1 2 5 . ?
muas 2 2 ? . 3
Cov neeg yuav khoom 3 ? 2 .
. . . . . . . .
Cov neeg yuav khoom 500,000 ? 2 . 1

Soj ntsuam thiab hackers nyob ib ncig ntawm lub ntiaj teb no twb kos rau cov kev sib tw, thiab los ntawm 2008 ntau tshaj li 30,000 tus neeg tau ua hauj lwm rau nws (Thompson 2008) . Nyob rau lub chav kawm ntawm lub contest, Netflix tau txais ntau tshaj li 40,000 npaj kev daws teeb meem los ntawm ntau tshaj 5,000 pawg (Netflix 2009) . Obviously, Netflix yuav tsis tau nyeem thiab to taub tag nrho cov tswv yim daws teeb meem. Qhov tseem khiav ntseeg nkaws, txawm li cas los, vim hais tias lub ntsiab yooj yim mus saib. Netflix yuav cia li muaj ib tug computer piv rau kwv yees kev mus rau lub nyob tawm kev los ntawm ib tug pre-teev metric (cov metric lawv siv yog lub square hauv paus ntawm qhov phem-squared kev ua yuam kev). Nws yog qhov no muaj peev xwm mus sai sai soj ntsuam kev daws teeb meem uas enabled Netflix mus txais cov ntsiab los ntawm txhua leej txhua tus, uas muab tau ib qho tseem ceeb vim hais tias zoo tswv yim los ntawm ib co ceeb qhov chaw. Nyob rau hauv qhov tseeb, lub yeej haum twb xa los ntawm ib pab neeg pib los ntawm peb cov kev soj ntsuam uas twb tsis muaj ua ntej kev lub tsev movie pom zoo systems (Bell, Koren, and Volinsky 2010) .

Ib tug zoo nkauj nam ntawm lub Netflix nqi zog yog tias nws enabled txhua leej txhua tus nyob rau hauv lub ntiaj teb no kom muaj lawv cov tshuaj soj ntsuam haum. Thaum cov neeg uploaded lawv kwv yees kev, lawv tsis yuav tsum upload lawv kawm ntawv, lawv lub hnub nyoog, haiv neeg, poj niam txiv, kev sib deev orientation, los yog dab tsi txog lawv tus kheej. Yog li ntawd, kwv yees kev ntawm ib tug naas ej xib fwb los ntawm Stanford tau kho raws nraim tib yam li cov neeg los ntawm ib tug tub hluas nyob rau hauv nws chav pw. Tu siab, qhov no yog tsis muaj tseeb nyob rau hauv feem ntau cov kev sib raug tshawb fawb. Uas yog, rau feem ntau cov kev sib raug tshawb fawb, kev ntsuam xyuas yog heev lub sij hawm siv thiab cov subjective. Yog li ntawd, feem ntau cov kev tshawb fawb cov tswv yim yog tsis tiag soj ntsuam, thiab thaum twg cov tswv yim yog soj ntsuam, nws yog ib qho nyuaj rau detach cov kev soj ntsuam los ntawm tus creator ntawm lub tswv yim. Vim hais tias kev daws teeb meem yog ib qho yooj yim mus saib, qhib hu pub soj ntsuam mus saib tau tag nrho cov muaj feem zoo kawg nkaus ntsiab uas yuav poob los ntawm cov kab nrib pleb yog hais tias lawv tsuas xam tau tias yog kev daws teeb meem los ntawm nto moo professors.

Piv txwv li, thaum ib tug taw tes thaum lub sij hawm Netflix nqi zog ib tug neeg uas lub screen npe Simon Funk muab tso rau nws blog ib tug npaj tov raws li nyob rau hauv ib tug zoo tshaj tus nqi decomposition, ib mus kom ze ntawm linear algebra uas twb tsis tau siv yav tas los los ntawm lwm yam koom. Funk tus blog ncej yog ib txhij kev thiab weirdly kev. Twb no blog ncej piav txog ib tug zoo tshuaj los yog nws ib tug pov tseg ntawm lub sij hawm? Nyob sab nraum ntawm ib tug qhib hu project, cov tshuaj yuav tsis tau txais loj kev ntsuam xyuas. Tom qab tag nrho Simon Funk tsis yog ib tug xib fwb ntawm Cal Tech los yog MIT; nws yog ib tug software tsim tawm uas, thaum lub sij hawm, twb ev hnab nce roob nyob ib ncig ntawm New Zealand (Piatetsky 2007) . Yog hais tias nws tau tuaj lub tswv yim no rau ib tug engineer ntawm Netflix, nws yuav luag yeej yuav tsis tau npaum li cas tiag.

Qhov zoo ces, vim hais tias qhov kev luj xyuas kev twb meej thiab yooj yim mus thov, nws kwv yees kev raug soj ntsuam, thiab nws yog instantly tseeb hais tias nws mus kom ze yog haib heev: nws rocketed plaub qhov chaw nyob rau hauv cov kev sib tw, ib tug zoo kawg li tshwm sim muab hais tias lwm pab pawg twb tau ua hauj lwm rau lub hlis rau qhov teeb meem. Nyob rau hauv lub kawg, qhov chaw ntawm Simon Funk tus mus kom ze tau siv los ntawm zoo tag nrho loj competitors (Bell, Koren, and Volinsky 2010) .

Qhov tseeb hais tias Simon Funk xaiv sau ib tug blog post dab nws mus kom ze, es ua kom nws tsis pub leejtwg paub, kuj tso siab hais tias muaj ntau cov neeg nyob rau hauv lub Netflix nqi zog twb tsis heev dua lwm yam kev mob siab los ntawm lub lab duas nqi zog. Es, ntau cov neeg kuj ciali yuav txaus siab rau kev txawj ntse sib tw thiab lub zej lub zos uas tsim nyob ib ncig ntawm qhov teeb meem (Thompson 2008) , kev xav uas kuv cia siab tias muaj ntau soj ntsuam ntawm muaj peev xwm to taub.

Lub Netflix nqi zog yog ib tug classic piv txwv ntawm ib qho kev qhib hu. Netflix nyiag ua ib lo lus nug uas muaj ib lub hom phiaj (cua movie ratings) thiab solicited daws los ntawm ntau tus neeg. Netflix muaj peev xwm los soj ntsuam tag nrho cov kev daws teeb meem vim hais tias lawv yooj yim kom paub tseeb tshaj los ua, thiab thaum kawg Netflix khaws qhov zoo tshaj plaws tshuaj. Tom ntej no, kuv mam li qhia koj yuav ua li cas no tib yam mus kom ze yuav siv tau nyob rau hauv ib thiab kev cai lij choj.