FastText Subword
FastText benn xeetu Facebook 2016 la buy wane baat bu nekk ni sac n-grams, suko defee mu mëna tabax vecteur doonte ci baat yi mu musul gis ci diiru tàggat.
Résumé
This subword approach excels at morphologically rich languages, typos, and rare words where Word2Vec and GloVe fail.
Plongeur bu xóot
FastText, bi Facebook AI Research defar (Bojanowski, Grave, Joulin, Mikolov) ci 2016, dafay yokk xeetu Skip-Gram ci xaaj baat bu nekk ci n-gram. Baatu "fu" ak n-gram ci guddaay 3 dafay nekk <wh, whe, her, ere, re> boole ci token baat bi yépp, fu parenthèses angle yi di màndargaal peggu baat yi. Vecteur u baat mooy limu vecteur n-gram yi ci nekk. Loolu dafay tekki ni FastText mën na defar vecteur ngir baat bu nekk ci bitti vocabulaire bu melni "gëmul" ci ay piyeesu subword yuñ miin, te dafay jàpp morphologie buñ bokk, kon "daw", "dawalkat", ak "dawal" dañuy lëkkaloo ci seen bopp. Benn projet bi dafay yónnee itam ab classificateur text lineaire bu gaaw te jaar yoon ("fastText" mode supervisé) ñu koy jëfandikoo ci liggéey yu melni ràññee làkk ak etiket ci anam wu yaatu.
Gis-gis xarala
Bépp araf n-gram dañu koy hash ci benn tablo bucket bu am dayo bu takku ba noppi ñu jox ko vecteur boppam; Representation bu benn baat mooy limu vecteur n-gram yi ci bokk, ñu tàggat leen ci benn mébetu Skip-Gram buy jël échantillonnage négatif ni Word2Vec. Séddoo parametru subword ci biir baat yi moo waral morphologie di toxal ak li waral baat yi ñu gisul ba leegi di am vecteur yu am xel. Classificateur bi ñuy saytu dafay jëfandikoo model sac-de-features bu noonu mel ak softmax hierarchique, muy gaaw lool ci CPU yi.
njeextalu pexe
Gaawaay ak yaatuwaay
Liggéeyukaay yi ci làkk yi mën nañu gëna gaaw te duñu yàq deggoo gi.
Dugg ak yegg
Dafay yaatal jëfandikoo gi ci làkk yi ak ci anam yi ñuy jokkoo.
dogal yu gëna leer
Ekip yi mën nañu gëna yàgg ci àtte ci jamono ji otomatisation di liggéey ci baamtu.
Ëlëgu FastText Subword
Xalaatu subword bu FastText dafa wane ni mooy gëna am solo: transformatër yu bees yi dañuy jëfandikoo pexe yu melni Encodage Byte-Pair ak tokenization WordPiece ngir mëna jëfandikoo bépp dugal bu amul benn baat buñ tëral. Facebook genne na ay vecteur FastText yuñ njëkka tàggat ngir 157 làkk, loolu moo tax mu nekk royuwaay ci NLP yu bari làkk yi ak yu néew doole yi, fu model yu yaatu yi mënu ñu jëfandikoo. Ginaaw model yu ndaw yi ci aparey yi ak ci boor yi dañu gëna am solo, emprent bu ndaw bu FastText ak gaawaayu CPU bi taxna mu am solo ci xaaj mbindu liggéey bi.
Doxal ci àdduna dëgg
Defar ay vecteur ngir ay baat yuñu juum wala yuñ musul gis lu ci melni "realy" wala tur produit yu bees
Vecteur yu ubbeeku yu Facebook yuñ tàggat bu njëkk, ñuy wax ci 157 làkk ngir seetlu ak etiketu làkk yu bari
Xamme làkk bu gaaw ak xaajale spam/topic ci CPU te amul GPU
Jëfandikoo làkk yu bari morphologie yu melni Finlande wala Turc fu baat yi di jël jëmm yu bari
Risk yi ak balustrade yi
Lépp lu jaarul yoon mën na dugg ci rapoor yi, jàppale ci liggéey bi, wala ci njariñu gëstu bi.
Sensibilite bu gaaw mën na jur njariñ yu wuute ci laajte yu noonu mel.
Done yu am solo mën nañu feeñ sudee seytu jëfandikoo gi néew doole.
Roadmap ngir samp gi
Mandargal formaa génne gi, melokaan bi, ak standard kalite yi laata ngay dugal ko.
Tontu yu am solo ak balluwaay yu wóor saa yu dëggu bi di am solo.
Fexeel am barabu xool nit ñi ngir am njariñ yu am solo.
Toppal anami gacce yi ak di faral di tàggataat ay laaj wala def-liggéey.
Weyal di banneexu
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 FastText Subword Embeddings 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
Gis bi ci topp
Matryoshka
Laaj yi ñuy faral di laaj
What is FastText Subword Embeddings?
FastText benn xeetu Facebook 2016 la buy wane baat bu nekk ni sac n-grams, suko defee mu mëna tabax vecteur doonte ci baat yi mu musul gis ci diiru tàggat. Xeetu subword bii dafa gëna baax ci làkk yu bari morphologie, njuumti ci bind, ak baat yu bariwul fu Word2Vec ak GloVe jàllul.
naka la FastText di màndargaale benn baat?
FastText dafay xaaj baat bu nekk ci arafu n-gram ba noppi boole seeni vecteur ngir sos misaalu baat bi.
Ban jafe-jafe bu mag la FastText mëna daan ci Word2Vec ak GloVe?
Ndax FastText dafay defar ay vecteur ci ay piyeesu subword, mën na tabax représentation ngir baat yi mu musul gis ci tàggat.
Ngir baat "fan" ak n-gram yu am 3 araf, ban n-gram mooy baax (ak ay màrker ci pegg yi)?
"fu" dafay joxe ay trigram yu melni <wh, fu, moom, balaa, re>, boole ci baat bi yépp; "whe" bokk na ci ñoom.
Ban mébetu tàggat yaram la modelu embedding bu FastText tabax ci kaw?
FastText dafay yokk Skip-Gram ak mébetu jël misaal bu baaxul, yokk ci vecteur n-gram yu ndaw.
Lan moo tax FastText am njariñ ci làkk yu melni Finlande wala Turc?
Làkk yi am morphologie yu bari dañuy defar xeeti baat yu bari; séddoo vecteur subword dafay tax FastText mëna leen boole ci seen bopp.