Maidowa Maimaitawa
Sake dawo da martaba shine mataki na biyu na bincike na zamani: bayan mai saurin dawo da shi ya ja saitin ɗan takara, ƙirar da ta fi ƙarfin ta sake ƙima waɗancan ƴan takarar don haka waɗanda suka dace da gaske sun tashi zuwa sama.
Dubawa
It is the quality boost behind better search and more accurate RAG systems.
Zurfafa nutsewa
Ƙirar da aka haɓaka da bincike da maidowa yawanci suna gudana cikin matakai biyu. Na farko, mai saurin dawo da sauri (BM25 na tushen keyword ko bincike mai zurfi) ya ɗauki babban wurin shakatawa na ɗan takara — faɗi saman 100 — yana haɓakawa don tunawa da sauri. Sa'an nan mai reranker ya bincika waɗannan 'yan takarar a hankali kuma ya sake yin odar su ta dacewa, yana inganta daidaitattun a saman. Na'urar reranker na al'ada shine mai rikodin giciye: yana ciyar da tambaya da kowane ɗan takara daftarin aiki tare zuwa na'urar wuta don haka hankali zai iya kwatanta su da kalma, yana samar da ma'auni guda ɗaya da suka dace. Wannan ya fi daidai fiye da abubuwan da aka haɗa masu zaman kansu na mai dawo da su amma kuma ya yi jinkirin yin aiki a kan gabaɗayan ƙungiyar-saboda haka ƙirar matakai biyu. A cikin RAG, sake fasalin mai kyau yana nufin samfurin yana ganin mafi dacewa wurare, rage hallucination da inganta ingancin amsa.
Fahimtar Fasaha
Maɓallin maɓalli shine mai-encoder tare da mai rikodin giciye. Mai rikodin rikodi guda biyu yana haɗa tambaya da daftarin aiki daban, don haka za'a iya ƙididdige ƙididdiga kuma a kwatanta su da samfuran ɗigo masu sauri-mai girma don dawo da matakin farko. Mai rikodin giciye yana haɗa tambaya da daftarin aiki kuma yana gudanar da su tare ta hanyar transfoma, yana ba da cikakkiyar kulawar kulawa da dacewa. Ƙirar-ƙirar ƙira ta fi daidai amma ba za su iya ƙididdige abubuwan daftarin aiki ba, don haka an keɓe su don sake tsara ƙaramin saitin ɗan takara maimakon bincika komai.
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 Maidowa Reranking
Reranking shine tsakiyar binciken samarwa da RAG, kuma kayan aikin yana faɗaɗa cikin sauri. APIs ɗin da aka karɓa (kamar Cohere Rerank) da buɗaɗɗen ƙira-ƙirar giciye sun sauƙaƙa kunnawa. Sabbin kwatance sun haɗa da yin amfani da manyan samfuran harshe da kansu a matsayin masu ba da lissafi waɗanda ke yin tunani kan kowane ɗan takara da aka saita lokaci guda, ƙirar haɗin gwiwa kamar ColBERT waɗanda ke daidaita saurin gudu da daidaito, da kuma koyon haɗakar masu dawo da yawa. Kamar yadda mahallin windows ke girma, yi tsammanin haɗaɗɗiyar maɗaukaki tsakanin sake tsarawa da kuma yadda ake zaɓin wurare da oda don tsarawa.
Aiwatar da Gaskiyar Duniya
A RAG chatbot ya dawo da sassa 50 tare da binciken vector, sannan mai rikodin giciye ya canza su don haka manyan 5 da aka ciyar da LLM sun fi dacewa.
Binciken rukunin yanar gizon e-kasuwanci yana amfani da BM25 don tunawa, sannan mai reranker ya sake yin odar samfura ta hanyar dacewa don ɗaga juzu'i.
Kira API rerank da aka shirya (misali, Cohere Rerank) don sake yin odar abubuwan nema ba tare da horar da ƙirar al'ada ba.
Yin amfani da ma'amala na ƙarshen salon ColBERT don sake sanya 'yan takara tare da daidaiton rikodin-kusa-ƙusa a ƙananan latency.
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 Retrieval Reranking 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
Ingancin Maidowa
Tambayoyin da ake yawan yi
What is Retrieval Reranking?
Sake dawo da martaba shine mataki na biyu na bincike na zamani: bayan mai saurin dawo da shi ya ja saitin ɗan takara, ƙirar da ta fi ƙarfin ta sake ƙima waɗancan ƴan takarar don haka waɗanda suka dace da gaske sun tashi zuwa sama. Yana da ingantacciyar haɓakawa a bayan mafi kyawun bincike da ingantaccen tsarin RAG.
Menene aikin mai sakewa a bututun mai?
Reranking shine mataki na biyu: a hankali yana sake yin odar ƴan takara mai saurin dawo da shi, yana inganta daidaito a saman.
Me ke bambanta mai rikodin giciye da mai rikodi?
Ketare-encoders suna haɗa tambaya da daftarin aiki kuma suna gudanar da su tare, suna ba da cikakkiyar kulawar giciye da daidaito mafi girma.
Me yasa ake amfani da maƙallan giciye don sakewa maimakon dawo da matakin farko?
Saboda tambayar da kowace takarda dole ne a sarrafa su tare, ba za ku iya yin ƙididdige abubuwan daftarin aiki ba, don haka maƙallan giciye kawai suna gudana akan ƙaramin saitin ɗan takara.
Menene mai dawo da matakin farko yakan inganta don?
Mai dawo da shi yana jefa raga mai fadi, mai sauri don ƙara yawan tunawa; mai jujjuyawar sannan yana kaifafa daidaito akan tafkin.
Ta yaya ingantaccen matsayi ke taimakawa tsarin RAG?
Ciyar da LLM mafi mahimmancin sassan yana inganta daidaiton amsa kuma yana yanke abubuwan da ake so.