Izinjini zokusesha zanamuhla ziguquke kakhulu ngoba azinikezwa kuphela ngokubuyisa amakhasi ngemuva kokuthola igama elilodwa noma amaningi, kepha futhi zizama ukuphendula imibuzo, ziphakamisa okuqukethwe, njll. Abasebenzisi banekhono lokusesha kwezinye izinto, njengezithombe.
Yebo kunjaloUkulandela okuncamelayo kokucinga kwabasebenzisi akukusha: Kube ngumzabalazo onzima selokhu kwaqala ukuseshwa kwewebhu.
Kepha manje, kuya ngokuya kuba nzima ukuhlangabezana nalezi zidingo eshintsha njalo, ngenxa yentuthuko yezobuchwepheshe bokufakelwa, kufaka phakathi lezo ezakhiwe yithimba labacwaningi be-Bing nabaphenyi beLicwaningo Lokucwaninga iMicrosoft.
"Ubuhlakani bokufakelwa benza imikhiqizo esisebenza nayo iye ngokwengeziwe futhi ibe yemvelo", URangan Majumder uthe, Umphathi Wezinhlelo zeQembu le-Bing Research ne-Artificial Intelligence ye-Microsoft.
Usebenzisa ama-vectors ukusesha okungcono
I-Los Ama-algorithm we-Bing wokufunda ngomshini asetshenziselwa ukudala ama-vectors, (empeleni ukumelwa kwedijithali kwegama, iphikseli yesithombe, noma elinye iphoyinti ledatha). I-vector kusiza ukuthatha lokho okushiwo ucezu lwedatha, kungaba umbhalo ekhasini lewebhu, izithombe, umsindo noma amavidiyo.
Lapho iphuzu lezinombolo selinikezwe idatha ezimisele, ingahlela noma imephu izithwali, kubekwe izinombolo eziseduze komunye nomunye ukumela ukufana. Le miphumela esondele ikhonjiswa kubasebenzisi, ithuthukisa imiphumela yosesho.
IMicrosoft yaqala ukugxila kuyo ubuchwepheshe obusekelwe kubo Ucwaningo lwevektha ye-Bing lapho onjiniyela benkampani beqala ukubona amaphethini angajwayelekile emaphethini wokusesha wabasebenzisi.
"Ngokuhlaziya omagazini bethu, ithimba lithole ukuthi imibuzo yokucinga ibiba mide futhi iba mide," kusho uMajumder.
Lokhu kuphakamise ukuthi abasebenzisi bebebuza imibuzo eminingi, benikeza imininingwane eminingi kakhulu ngenxa yemiphumela edlule ebingagculisi ngokuseshwa kwegama elingukhiye, noma "ukuzama ukwenza njengamakhompiyutha" lapho bechaza izinto ezingabonakali.
Ngosesho lwe-Bing, ukukhishwa kwe-vectorization kuye kwanda kwaba ngaphezu kwedatha enezikhombo ezingaphezu kwezigidigidi eziyi-150 izinjini zokusesha ukwenza ngcono ukuqhathanisa ngamagama angukhiye wendabuko.
Lezi faka amagama alula, izinhlamvu, amazwibela wekhasi lewebhu, imibuzo ephelele, nokunye abezindaba. Lapho umsebenzisi enze ukusesha, i-Bing ingahlaziya ama-vectors anezikhombo futhi inikeze ukufana okuhle kakhulu.
Ukwenza imephu kweVector kubunjwe kusetshenziswa ubuchwepheshe bokufunda obujulile ukuze kwenziwe ngcono okuqhubekayo.
Amamodeli abheka okokufaka okufana nokuchofoza kwabasebenzisi bokugcina ngemuva kosesho ukuze baqonde kangcono incazelo yalolu sesho.
I-Space Partition Tree neGrafu ikhishwe njengomthombo ovulekile
Eqinisweni, IMicrosoft isebenzisa i-algorithm ebizwa ngeSpace Partition Tree and Graph (SPTAG). Umbuzo wokufaka uguqulelwa ku-vector, futhi i-SPTAG isetshenziselwa ukuthola ngokushesha "omakhelwane abaseduze," okungukuthi, ama-vectors afana nokufakwayo.
"IMicrosoft isebenzisa ukusesha i-vector ngenjini yayo yokusesha ye-Bing, ubuchwepheshe obusiza i-Bing ukuthi iqonde kangcono inhloso yokusesha ngezigidigidi zewebhu futhi ithole imiphumela efanelekile evela ezigidini zamakhasi e-web."
Microsoft yenze kwatholakala kuwo wonke umuntu, njengephrojekthi yomthombo ovulekile ku-GitHub,
"Elinye lamathuluzi obuhlakani bokuthuthuka asezingeni eliphezulu futhi afaneleke kakhulu ukuhlangabezana nezidingo zokusesha ezihlala zikhona zabasebenzisi."
NgoLwesithathu, umshicileli uphinde wathumela amasampula amasu womsebenzisi kanye nevidiyo ehambisana nalawa mathuluzi ngelebhu yezobunhloli yokufakelwa kaMicrosoft.
Ithimba le-Bing lithembile ukuthi banethemba lokuthi umnikelo ovulekile womthombo ungasetshenziswa ngezinkampani eziholayo noma izinhlelo zokusebenza ukukhomba ulimi olukhulunywayo olususelwa kumazwibela alalelwayo, noma kumasevisi afaka phakathi izithombe eziningi, njengohlelo lokusebenza oluvumela abasebenzisi ukufanisa idatha nosesho.