Ibice byihuta bya Subword
FastText nuburyo bwa Facebook AI 2016 bwerekana buri jambo nkumufuka wimiterere n-garama, bityo irashobora kubaka ibice ndetse no kumagambo itigeze ibona mugihe cyamahugurwa.
Incamake
This subword approach excels at morphologically rich languages, typos, and rare words where Word2Vec and GloVe fail.
Kwibira cyane
FastText, yakozwe na Facebook AI Ubushakashatsi (Bojanowski, Imva, Joulin, Mikolov) mu 2016, yagura icyitegererezo cya Skip-Gram mu guca buri jambo mu nyuguti n-garama. Ijambo "aho" hamwe n-garama z'uburebure bwa 3 rihinduka <wh, whe, her, ere, re> wongeyeho ijambo ryuzuye ikimenyetso, aho inguni zinguni zerekana imipaka. Ijambo ryijambo nigiteranyo cya n-garama yacyo. Ibi bivuze ko FastText ishobora guhimba vector yijambo ritari mu magambo nka "kutizera" uhereye kumagambo asanzwe amenyerewe, kandi ifata morphologie isangiwe, bityo "kwiruka," "kwiruka," na "kwiruka" bifitanye isano bisanzwe. Umushinga umwe kandi wohereza umurongo wihuse, wuzuye umurongo wanditse ("yihuta" igenzurwa) ikoreshwa kumirimo nko kumenyekanisha ururimi no gutondekanya murwego runini.
Ubushishozi
Buri nyuguti n-garama yogejwe mumeza yindobo ihamye kandi igenera vector yayo; ijambo ryerekana ni igiteranyo cyibigize n-garama, byatojwe hamwe na sample-sample Skip-Gram intego nka Word2Vec. Uku kugabana ibipimo byibisobanuro mumagambo niyo mpamvu morphologie yimura nimpamvu amagambo atagaragara agifite ibice byumvikana. Kugenzura ibyiciro bikoresha imifuka isa-yerekana ibintu hamwe na softmax ya hierarchique, bigatuma yihuta cyane kuri CPU.
Ingaruka z'Ingamba
Umuvuduko n'igipimo
Ururimi rwakazi rushobora kugenda byihuse nta gutamba guhuzagurika.
Kugera no kugera
Yagura uburyo bwindimi nuburyo bwo gutumanaho.
Ibyemezo bisobanutse
Amakipe arashobora kumara umwanya munini murubanza mugihe automatike ikora gusubiramo.
Ejo hazaza hihuta byihuta Subword
Igitekerezo cyibisobanuro byihuse bya FastText byagaragaye ko ari ishingiro: abahinduzi ba kijyambere bakoresha tekinike zijyanye na Byte-Pair Encoding na WordPiece tokenisation kugirango bakemure ibyinjijwe byose nta magambo ahamye. Facebook yasohoye ibice byihuta byihuta byindimi 157, bikomeza kuba ishingiro ryindimi nyinshi kandi zifite amikoro make NLP aho moderi nini zidakwiye. Nka ntoya ku gikoresho na moderi yicyitegererezo bigira akamaro, Ikirenge gito cya FastText hamwe na CPU umuvuduko ukomeza kuba ingirakamaro kubikorwa byo gutondekanya inyandiko.
Gushyira mu bikorwa Isi
Gukora ibice byamagambo yanditswe nabi cyangwa atigeze aboneka nka "realy" cyangwa amazina y'ibicuruzwa bishya
Facebook ifungura-isoko-yerekana ibice 157 byo gushakisha indimi nyinshi
Kumenyekanisha ururimi rwihuta na spam / ingingo itondekanya kuri CPU idafite GPU
Gukemura indimi zikungahaye kuri morfologiya nka finine cyangwa Turukiya aho amagambo afata uburyo bwinshi bwuzuye
Ingaruka & Kurinda
Ibintu bifatika bishobora kwinjiza bucece raporo, gushyigikira imigendekere, cyangwa ibisubizo byubushakashatsi.
Kwihuta byihuse birashobora gukora ibisubizo bidahuye mubisabwa bisa.
Ibyanditswe byumvikana birashobora kugaragara niba kugenzura kugenzura ari ntege.
Igishushanyo mbonera
Sobanura imiterere isohoka, amajwi, hamwe nubuziranenge mbere yo gutangira.
Ibisubizo byibanze hamwe nisoko yizewe igihe cyose ukuri kwingirakamaro.
Komeza kugenzura abantu kugenzura ibisubizo byinshi.
Kurikirana uburyo bwo kunanirwa no kongera imyitozo cyangwa akazi gahoraho.
Komeza Ubushakashatsi
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
Ubuyobozi bukurikira
Ibyapa bya Matryoshka
Ibibazo bikunze kubazwa
What is FastText Subword Embeddings?
FastText nuburyo bwa Facebook AI 2016 bwerekana buri jambo nkumufuka wimiterere n-garama, bityo irashobora kubaka ibice ndetse no kumagambo itigeze ibona mugihe cyamahugurwa. Ubu buryo bwibanze bwibanze ku ndimi zikungahaye kuri morfologiya, amakosa yamagambo, n'amagambo adasanzwe aho Word2Vec na GloVe binaniwe.
Nigute FastText yerekana ijambo rimwe?
FastText ibora buri jambo mumiterere n-garama kandi igateranya ibice byayo kugirango ijambo rihagarare.
Ni izihe mbogamizi zikomeye za Word2Vec na GloVe yihuta?
Kuberako FastText igizwe na vectors kuva ibice byijambo, irashobora kubaka ishusho yamagambo itigeze ibona mumahugurwa.
Kubwijambo "aho" hamwe ninyuguti 3 n-garama, niyihe n-garama yemewe (hamwe nibimenyetso byimbibi)?
"aho" itanga trigramma nka <wh, whe, her, ere, re>, wongeyeho ijambo ryuzuye ikimenyetso; "ibiziga" ni kimwe muri byo.
Niyihe ntego yibanze yo guhugura icyitegererezo cya FastText yubakiyeho?
FastText yagura Skip-Gram hamwe nintego mbi yo gutoranya, wongeyeho ijambo n-garama.
Ni ukubera iki FastText ari ingirakamaro cyane mu ndimi nka Finilande cyangwa Turukiya?
Indimi zikungahaye kuri morfologiya zitanga amagambo menshi; kugabana subword vectors ireka FastText ibahuza muburyo busanzwe.