Hukunce-hukuncen Jumla-BERT
Jumla-BERT (SBERT) tana daidaita BERT don samar da tsayayyen tsayin tsayi guda ɗaya don jimla gabaɗaya, don haka ana iya kwatanta ma'ana da kamancen cosine mai sauri.
Dubawa
It made semantic search and clustering over millions of sentences practical, turning a job that took BERT hours into milliseconds.
Zurfafa nutsewa
A fili BERT na iya kwatanta jimloli biyu don kamance, amma ta hanyar ciyar da duka biyu tare ta hanyar hanyar sadarwa, wanda ke da jinkiri sosai a sikelin: kwatanta jimloli 10,000 bi da bi zai buƙaci kusan wucewa miliyan 50. Jumla-BERT, wanda Reimers da Gurevych suka gabatar a cikin 2019, yana gyara wannan ta hanyar amfani da hanyar sadarwa ta siamese (twin): Hasumiya ta BERT guda biyu tare da ma'aunin ma'auni guda ɗaya kowannensu yana ɓoye jumla ɗaya da kansa, sannan matakin haɗaɗɗiya (yawanci yana nufin haɗawa kan abubuwan haɗawa) yana haifar da vector ɗaya a cikin jumla. An daidaita ƙirar ta yadda jimlolin kamanceceniya na ma'ana su zo kusa da juna a sararin samaniya. Yanzu kowace jumla an sanya su sau ɗaya a cikin abin da za a sake amfani da shi, kuma kamanni ya zama samfurin digo mai arha, yana ba da damar bincike, cirewa, da tari a ma'auni mai girma.
Fahimtar Fasaha
SBERT yawanci ana horar da shi tare da gine-ginen siamese da maƙasudin bambanci ko uku. Bayanai na Harshen Halitta na gama gari: an haɗa nau'i-nau'i na haɗe tare, an ture sabani. Hasumiyai biyu suna raba ma'auni, don haka shigar da su daidai ne. Ma'anar haɗawa a kan na'urorin alama na ƙarshe gabaɗaya ya fi yin amfani da alamar [CLS] kaɗai, yana samar da abubuwan haɗawa inda kamannin cosine ke dogaro da kusancin ma'ana.
Dabarun Tasiri
Gudu da sikelin
Gudun aikin harshe na iya tafiya da sauri ba tare da sadaukar da daidaito ba.
Shiga ku isa
Yana faɗaɗa damar shiga cikin harsuna da salon sadarwa.
Shawarwari masu haske
Ƙungiyoyi za su iya ciyar da ƙarin lokaci akan hukunci yayin da aiki da kai ke sarrafa maimaitawa.
Makomar Hukunce-hukuncen Jumla-BERT
SBERT-style bi-encoders yanzu suna ƙarfafa tsarar da aka haɓaka, suna ciyar da mahallin da ya dace ga manyan nau'ikan harshe. Filin yana tafiya zuwa mafi girma na koyarwa-saukar nau'ikan haɗaɗɗiyar haɗaɗɗiya, haɗaɗɗen harsuna da yawa da na zamani, da wakilcin Matryoshka waɗanda za a iya yanke girman su don saurin gudu. Haɗaɗɗen bututun sun haɗa da maido da rikodi cikin sauri tare da sake saita mai rikodin giciye a hankali, haɗa ma'aunin SBERT tare da daidaito mafi girma akan manyan 'yan takara.
Aiwatar da Gaskiyar Duniya
Injunan bincike na Semantic sun haɗa tambaya da duk takaddun, sannan su dawo da mafi kusa da vector maimakon dogaro da maɓalli na maɓalli.
Tsarukan haɓaka-dawowa suna amfani da abubuwan sakawa na SBERT don ɗora hanyoyin da suka dace don ƙaddamar da amsoshin chatbot.
Kayan aikin tallafi na abokin ciniki suna tattara tikiti masu shigowa ta hanyar haɗa kamanni zuwa kwafi ko batutuwa masu alaƙa ta atomatik.
Laburaren Python-masu jumlolin jumla suna ba da ƙirar SBERT da aka riga aka horar don fassarori ma'adinai da kuma cire rubutun kusa da iri ɗaya.
Hatsari & Tsare-tsare
Abubuwan da aka ruɗe suna iya shigar da rahotanni cikin nutsuwa, kwararar tallafi, ko abubuwan bincike.
Hankali na gaggawa na iya ƙirƙirar sakamako mara daidaituwa a cikin buƙatun iri ɗaya.
Za a iya fallasa bayanan rubutu mai ma'ana idan ikon samun dama yana da rauni.
Taswirar Hanya
Ƙayyade tsarin fitarwa, sautin, da ma'auni masu inganci kafin fitowa.
Amsa a ƙasa tare da amintattun tushe a duk lokacin da daidaito ya shafi mahimmanci.
Ajiye wurin binciken ɗan adam don abubuwan da ake samu masu girma.
Bibiyar tsarin gazawar kuma sake horar da tsokaci ko tafiyar aiki akai-akai.
Ci gaba da Bincike
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 Sentence-BERT 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
Jagora na gaba
Hasashen Hasashen Hasashen HyDE
Tambayoyin da ake yawan yi
What is Sentence-BERT Embeddings?
Jumla-BERT (SBERT) tana daidaita BERT don samar da tsayayyen tsayin tsayi guda ɗaya don jimla gabaɗaya, don haka ana iya kwatanta ma'ana da kamancen cosine mai sauri. Ya sanya bincike na ma'ana da tari sama da miliyoyin jimloli masu amfani, yana mai da aikin da ya ɗauki awoyi BERT zuwa millise seconds.
Wace babbar matsala tare da BERT bayyananne ke magance Sentence-BERT?
A fili BERT dole ne ya aiwatar da nau'ikan jimloli tare, don haka babban sikelin na biyu ba zai yuwu ba. SBERT tana ɓata kowane jumla sau ɗaya a cikin wani nau'i mai maimaitawa.
Wane tsarin gine-ginen cibiyar sadarwa ke amfani da Sentence-BERT?
SBERT tana amfani da tsarin siamese (tagwaye) inda masu rikodin BERT guda biyu ke raba ma'auni kuma kowannensu ya haɗa jumla ɗaya da kansa.
Wace dabarar haɗawa ce aka fi ba da shawarar don haɗawa da SBERT?
Ma'anar haɗawa sama da na'urori na ƙarshe gabaɗaya ya fi yin amfani da alamar [CLS] kaɗai don shigar da jumla.
Da zarar an shigar da jimloli, ta yaya ake auna kamanceninta?
Gabaɗayan batu na SBERT shine kamancen yana raguwa zuwa kwatancen samfurin cosine/dige-dige mai arha tsakanin abubuwan da aka riga aka lissafa.
Wani nau'in bayanan da aka saba amfani dashi don daidaita SBERT?
Ana amfani da bayanan NLI ko'ina: an haɗa nau'i-nau'i nau'i-nau'i tare kuma an ture saɓani, suna tsara sararin haɗawa mai ma'ana.