The Netflix Nrite-eji oghe oku na-aga ịkọ fim ndị mmadụ ga-amasị.
The kasị ọma mara emeghe oku na-aga oru ngo bụ Netflix Nrite. Netflix bụ online nkiri mgbazinye ụlọ ọrụ, na na 2000 ọ ulo oru Cinematch, a ọrụ ka nwere ike ikwu ihe nkiri na ahịa. Dị ka ihe atụ, Cinematch nwere ike na-achọpụta na ị mmasị Star Wars na Alaeze Ukwu etiwapụ Back mgbe ahụ nwere ike ikwu na ị na-ele Nlaghachi nke Jedi. Ná mmalite, Cinematch rụrụ ọrụ agụghị oké. Ma, n'elu N'ezie nke ọtụtụ afọ, Cinematch nọgidere na-amụtakwu ike ịkọ ihe na fim ahịa ga-enwe. Site 2006, Otú ọ dị, ọganihu na Cinematch plateaued. The eme nnyocha na Netflix gbalịrị mara mma nke ukwuu ihe nile ha nwere ike na-eche nke, ma n'otu oge, ha enyo enyo na e nwere ndị ọzọ echiche ndị na-enyere ha aka mma ha usoro. N'ihi ya, ha wee na ihe, mgbe ahụ, a buu ngwọta: na-emeghe oku na-aga.
Critical na omume ịga nke ọma nke Netflix Nrite bụ otú ahụ na-emeghe oku e mere, nke a haziri nwere mkpa maka otú emeghe oku ga-eji na-elekọta mmadụ nnyocha. Netflix mere abụghị naanị wepụta ihe unstructured arịrịọ maka echiche, nke bụ ihe ọtụtụ ndị na-eche na mgbe ha buru ụzọ tụlee ihe na-emeghe oku na-aga. Kama nke ahụ, Netflix jụrụ a doo anya nsogbu na a mfe nwale ibiere: ha mara ndị na-eji a set nke 100 nde nkiri Anyela ịkọ 3 nde ẹkenịmde-esi ratings (ratings na ọrụ mere kama na Netflix mere hapụ). Onye ọ bụla nwere ike ịmepụta otu algọridim na ike ịkọ 3 nde ẹkenịmde-esi ratings 10% mma karịa Cinematch ga-emeri 1 nde dollar. Nke a doro anya na mfe ime nwale ibiere-atụnyere buru amụma ratings ka ẹkenịmde-esi ratings-ekwu bụ na ndị Netflix Nrite e gosikwa dị otú ahụ n'ụzọ na ngwọta na-adị mfe iji lelee karịa n'ịwa; ọ tụgharịa ihe ịma aka nke rụọ Cinematch n'ime a nsogbu adabara na-emeghe oku na-aga.
Na October 2006, Netflix tọhapụrụ a dataset nwere 100 nde nkiri Anyela si banyere banyere 500,000 ahịa (anyị ga-atụle nzuzo nke a pụtara data ntọhapụ n'Isi nke 6). The Netflix data nwere ike conceptualized dị ka a nnukwu matriks na bụ mkpokọta 500,000 ahịa site na 20,000 na fim. Na n'ime ya ka matriks, e nwere ihe dị ka 100 nde ratings on a ọnụ ọgụgụ site 1 5 kpakpando (Isiokwu 5.2). Ihe ịma aka bụ iji hụrụ data na matriks ikwu ihe ga-3 nde ẹkenịmde-esi ratings.
Movie 1 | Movie 2 | Movie 3 | . . . | Movie 20,000 | |
---|---|---|---|---|---|
ahịa 1 | 2 | 5 | . | ? | |
ahịa 2 | 2 | ? | . | 3 | |
ahịa 3 | ? | 2 | . | ||
. . . | . | . | . | . | . |
ahịa 500,000 | ? | 2 | . | 1 |
Researchers na hackers gburugburu ụwa ji bụrụ ndị ihe ịma aka, ma site na 2008 ihe karịrị 30,000 mmadụ nọ na-arụ ọrụ na ya (Thompson 2008) . N'elu N'ezie nke zoo, Netflix natara ihe karịrị 40,000 chọrọ ihe ngwọta site n'ebe karịrị 5,000 ìgwè (Netflix 2009) . O doro anya na Netflix enweghị ike ịgụ ma ghọta ihe nile ndị a chọrọ ihe ngwọta. Dum ihe gbara ọsọ were were, Otú ọ dị, n'ihi na ngwọta dị mfe elele. Netflix ike dị nnọọ nwere kọmputa jiri ndị buru amụma ratings ka ẹkenịmde-esi ratings site a tupu kpọmkwem metric (akpan akpan metric ha na-eji bụ square mgbọrọgwụ nke pụtara-nhata njehie). Ọ bụ nke a ike ngwa ngwa inwale ngwọta nke nyeere Netflix na-anabata ngwọta si onye ọ bụla, nke mesịrị bụrụ mkpa n'ihi na ezi echiche wee site na ụfọdụ ihe ijuanya ebe. N'ezie, emeri ngwọta e osụk site otu ìgwè malitere site atọ na-eme nnyocha na amaghị ụlọ nkiri nkwanye usoro (Bell, Koren, and Volinsky 2010) .
One mara mma akụkụ nke Netflix Nrite bụ na ọ na-nyeere onye ọ bụla n'ụwa na-enwe ha ngwọta inyocha pụtara. Mgbe ndị mmadụ na uploaded ha buru amụma ratings, ha mkpa bulite ha agụmakwụkwọ nzere, n'afọ ndụ ha, agbụrụ, okike, mmasị ịkwa iko, ma ọ bụ ihe ọ bụla banyere onwe ha. N'ihi ya, e buru amụma ratings nke a ma ama prọfesọ si Stanford-emeso na-kpọmkwem otu ihe ahụ dị ka ndị si a afọ iri na ụma na ya ime. O di nwute na, nke a abụghị eziokwu na ọtụtụ na-elekọta mmadụ nnyocha. Ya bụ, n'ihi na ọtụtụ na-elekọta mmadụ research, nwale bụ nnọọ na-ewe oge na n'omume nke onye. Ya mere, ọtụtụ nnyocha echiche na-mgbe kpọrọ inyocha, na mgbe echiche na-Enyochala, ọ bụ ike ikwụpu ndị nwete si onye kere echiche. N'ihi na ihe ngwọta ndị mfe elele, na-emeghe oku-ekwe ka na-eme nnyocha iji nweta niile nwere ihe magburu onwe ngwọta nke ga-ada site cracks ma ọ bụrụ na ha na-atụle ihe ngwọta si a ma ama ndị ọkachamara.
Dị ka ihe atụ, n'otu oge n'oge Netflix Nrite onye na ihuenyo aha Simon Funk gwa ya blog a chọrọ ihe ngwọta dabere na a erikwa uru decomposition, na-abịaru nso site linear algebra a na-eji na mbụ site ọzọ sonyere. Funk si blog post bụ n'out oge oru na weirdly n'ịgba. E na nke a blog post na-akọwa ezi ihe ngwọta ka ọ bụ igbu oge? -Abụghị nke na-emeghe oku na-aga oru ngo, ihe ngwọta agaghịkwa ama natara nlezianya tụlee. Mgbe niile Simon Funk bụghị a prọfesọ na Cal Tech ma ọ bụ MIT; ọ bụ a software Mmepụta onye, n'oge ahụ, e backpacking gburugburu New Zealand (Piatetsky 2007) . Ọ bụrụ na o emailed echiche a na engineer na Netflix, ọ fọrọ nke nta N'ezie, a garaghị kpọrọ ihe.
Ọ dabara nke ọma, n'ihi na nwale ibiere doro anya na mfe ime, ya buru amụma ratings e inyocha, na ọ bụ n'otu ntabi anya na ya obibia bụ nnọọ dị ike: o rocketed ka ano ha ebe na mpi, a ezigbo N'ihi nyere na ndị ọzọ ìgwè nọworị -arụ ọrụ maka ọnwa na nsogbu. N'ikpeazụ, akụkụ nke Simon Funk si ejekwuru ji mee nta niile dị oké njọ asọmpi (Bell, Koren, and Volinsky 2010) .
Eziokwu ahụ bụ na Simon Funk họọrọ dee a blog post na-akọwa ya obibia, kama ịnwa na-ya na nzuzo, na-egosi na ọtụtụ ndị na-Netflix Nrite na-adịghị nanị na-akwali site nde dollar na-agbata n'ọsọ. Kama nke ahụ, ọtụtụ ndị keere òkè na-yiri na-enwe ndị nwere ọgụgụ isi ihe ịma aka na ndị obodo na mepụtara gburugburu nsogbu (Thompson 2008) , otú obi na m na-atụ anya ọtụtụ ndị nnyocha pụrụ ịghọta.
The Netflix Nrite bụ nnọọ ihe atụ nke na-emeghe oku na-aga. Netflix jụrụ otu ajụjụ na otu ihe mgbaru ọsọ (ịkọ nkiri Anyela) na-achọ ngwọta si ọtụtụ ndị. Netflix ike inwale a niile ngwọta n'ihi na ha mfe iji nyochaa karịa ike, n'ikpeazụ, Netflix ẹmende ihe ngwọta kasị mma. Next, m ga-egosi gị otú a emekwa otu ihe ahụ nwere ike ji mee bayoloji na iwu.