Word2Vec Skip-gram na CBOW
Word2Vec bụ usoro 2013 sitere na Google nke na-amụta vectors okwu siri ike site n'ịkọ okwu sitere na ndị agbata obi ha, na-atụgharị asụsụ ka ọ bụrụ geometry ebe okwu ndị yiri ya na-anọdụ ala nso.
Nchịkọta
It made the famous "king - man + woman ≈ queen" analogy possible and kicked off the modern embedding era.
Ime miri emi
Word2Vec, nke Tomas Mikolov na ndị ọrụ ibe webatara na Google na 2013, na-amụta vector (nke na-abụkarị ọnụọgụ 100-300) maka okwu ọ bụla site n'ịzụ netwọk neural nke nwere oyi akwa abụọ na-emighị emi na windo ọnọdụ na-amị amị. Ọ na-abịa na ụtọ abụọ. CBOW (Akpa Okwu na-aga n'ihu) na-ewere okwu ndị gbara ya gburugburu wee buru amụma okwu etiti na-efu efu, na-emekọ ihe ndị gbara ya gburugburu ọnụ. Skip-gram tụgharịrị nke a: ọ na-ewe etiti okwu wee nwaa ịkọ okwu gbara ya gburugburu. Ihe nlereanya ahụ adịghị eche banyere ọrụ amụma n'onwe ya; ihe mgbaru ọsọ bụ matriks arọ ọ na-amụta n'ụzọ, nke ahịrị ya na-aghọ okwu vectors. Okwu ndị na-apụta na ọnọdụ ndị yiri ya na-ejedebe na vector ndị yiri ya, na-ewepụta ihe ọ pụtara naanị site na nsonye.
Nghọta nka nka
Ịzụ softmax zuru oke n'elu nnukwu okwu anaghị adị ngwa ngwa, yabụ Word2Vec na-eji aghụghọ dị ka nlele na-adịghị mma, nke na-emegharị amụma dị ka ọnụọgụ ọnụọgụ abụọ: ịmata ọdịiche dị n'okwu gbara ya gburugburu site na obere okwu “adịghị mma”. Ọ na-esetịpụkwa mkpụrụokwu ugboro ugboro dị ka "nke ahụ" ma na-eji nkesa unigram-welitere-na-0.75 iji họrọ adịghị mma. CBOW dị ngwa ngwa ma dị mma maka okwu ugboro ugboro; Skip-gram nwere nlele na-adịghị mma na-eji okwu ndị na-adịghị ahụkebe na obere corpora mma.
Mmetụta atụmatụ
Ọsọ na ọnụ ọgụgụ
Usoro ọrụ asụsụ nwere ike ịga ngwa ngwa n'achụghị nkwụsi ike.
Nweta na iru
Ọ na-agbasawanye ohere n'ofe asụsụ na ụdị nzikọrịta ozi.
Mkpebi doro anya
Ndị otu nwere ike itinyekwu oge na ikpe ebe akpaaka na-ejikwa nkwughachi.
Ọdịnihu nke Word2Vec Skip-gram na CBOW
Edebere ihe ntinye dị ka Word2Vec n'ụzọ dị ukwuu site na ụdị ọnọdụ (ELMo, BERT, transformers) na-enye okwu vector dị iche iche dabere na nkebi ahịrịokwu, na-edozi nsogbu polysemy ebe "ụlọ akụ" nwere otu vector edoziri. Ma Word2Vec na-atachi obi ebe ọsọ, mfe, na nkọwa okwu: usoro nkwanye, ọchụchọ na dị ka ntọala nkuzi. Echiche ya bụ isi, nke pụtara na-apụta site na ọnụ ọgụgụ ndị na-emekọ ihe, ka bụ ihe ndabere echiche nke ụdị asụsụ ọgbara ọhụrụ niile.
Mmejuputa n'ezie n'ụwa
Spotify na Airbnb haziri Skip-Gram ịmụta ntinye nke egwu na ndepụta ("item2vec") site na usoro nnọkọ onye ọrụ maka ntụnye.
Na-eme ka nchọgharị ọmụmụ ihe na mgbasawanye otu okwu ka ajuju maka "laptọọpụ" na-ekwalite "akwụkwọ ndetu" na "kọmputa"
Ịchọta ntụnyere na mmekọrịta dị n'ederede, dị ka ụzọ abụọ isi obodo (Paris bụ France ka Tokyo dị na Japan)
Ịmalite akwa ntinye ntinye nke pipeline NLP buru ibu maka nyocha mmetụta na nhazi akwụkwọ na oke data
Ihe ize ndụ & okporo ụzọ nche
Eziokwu ndị e chepụtara echepụta nwere ike jiri nwayọ tinye akụkọ, nkwado nkwado, ma ọ bụ nsonaazụ nyocha.
Mmetụta ngwa ngwa nwere ike ịmepụta nsonaazụ na-ekwekọghị ekwekọ n'ofe arịrịọ ndị yiri ya.
Enwere ike ikpughe data ederede nwere mmetụta ma ọ bụrụ na njikwa ohere adịghị ike.
Map mmejuputa
Kọwaa usoro mmepụta, ụda, na ụkpụrụ ịdịmma tupu ibugharị.
Weghachite nzaghachi site na isi mmalite ntụkwasị obi mgbe ọ bụla izi ezi dị mkpa.
Debe ebe nleba anya mmadụ maka mpụta dị elu.
Sochie ụkpụrụ ọdịda ma na-azụghachi mkpali ma ọ bụ usoro ọrụ mgbe niile.
Nọgide na-eme nchọpụta
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Word2Vec Skip-Gram and CBOW quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ntuziaka na-esote
Netwọk okporo ụzọ wee gaa njikọ
Ajụjụ a na-ajụkarị
What is Word2Vec Skip-Gram and CBOW?
Word2Vec bụ usoro 2013 sitere na Google nke na-amụta vectors okwu siri ike site n'ịkọ okwu sitere na ndị agbata obi ha, na-atụgharị asụsụ ka ọ bụrụ geometry ebe okwu ndị yiri ya na-anọdụ ala nso. O mere ka ihe atụ a ma ama "eze - nwoke + nwanyị ≈ eze nwanyị" kwe omume wee malite oge ntinye nke oge a.
Kedu ihe Skip-gram architecture na-ebu amụma?
Skip-gram na-ewere otu okwu etiti wee nwaa ịkọ amụma nke ọ bụla gbara ya gburugburu, nke na-abụghị nke CBOW.
Kedu ihe bụ nsonaazụ bara uru nke ịzụ ụdị Word2Vec?
Ọrụ amụma bụ naanị ụzọ isi kwụsị; ebumnuche bụ matriks arọ amụtara nke ahịrị ya na-aghọ nnukwu okwu agbakwunyere.
Kedu ihe kpatara Word2Vec ji eji nlele adịghị mma?
Nlele na-adịghị mma na-edozi nsogbu ahụ dị ka nhazi ọnụọgụ abụọ megide mkpụrụokwu ole na ole, na-ezere softmax dị oke ọnụ karịa iri puku kwuru iri puku okwu.
Kedu ụdị Word2Vec na-arụkarị ọrụ nke ọma na okwu ndị na-adịghị ahụkebe na obere datasets?
Skip-gram nwere nlele na-adịghị mma na-achọ ijide okwu ndị na-adịghị ahụkebe nke ọma, ebe CBOW na-adị ngwa ngwa ma na-akwado okwu ugboro ugboro.
Kedu ihe ama ama nke Word2Vec vectors nke “eze - nwoke + nwanyị ≈ eze nwanyị” gosipụtara?
Vectors Word2Vec na-edobe mmekọrịta ka e wee dozie ihe atụ gbasara ọmụmụ site na mgbakwunye na mwepu vector dị mfe.