2.4.2 Forecasting kunye nowcasting

Ukuqikelela elizayo nzima, kodwa ukuqikelela langoku kulula.

Isicwangciso yesibini eziphambili ezisetyenziswa abaphandi kunye data lokuqwalasela uqikelelo. Ukuqikelela ikamva kunzima, kodwa kunokuba kakhulu kwinguqu abenzi bezigqibo, nokuba basebenza ziinkampani okanye oorhulumente.

Kleinberg et al. (2015) inika amabali amabini ukucacisa ukubaluleka nolwemo- kwiingxaki ezithile nkqubo. Khawube kumenzi enye nkqubo, ndiza kubiza uAnna yakhe, ojamelene imbalela kwaye kufuneka isigqibo sokuba ukuqesha shaman ukwenza umdaniso imvula ukwandisa amathuba imvula. Enye maker nkqubo, ndiya kuye kubiza Bob, kufuneka enze isigqibo sokuba athathe iambrela ukusebenza ukuphepha manzi endleleni egodukayo. Zombini Anna kunye Bob ungenza isigqibo ngcono ukuba baqonde imozulu, kodwa kufuneka bazi ukuba izinto ezahlukeneyo. Anna kufuneka aqonde ukuba umdaniso imvula ebangela imvula. Bob, kwelinye icala, akuyomfuneko ukuba nokuqonda nto kse; kufuneka nje kubekho echanileyo. Abaphandi Social badla kugxininisa koko Kleinberg et al. (2015) kubiza 'imvula umdaniso-like "iingxaki-ezo nkqubo zijolise kse-kwaye ngoyaba iingxaki nkqubo" umbrella-like "ezijolise lwemozulu.

Ndingathanda ukujolisa, Nangona kunjalo, uhlobo olukhethekileyo nolwemo- ebizwa nowcasting -oko elide efunyenwe yokudibanisa "ngoku" kunye "nokuqikelelwa." Kunokuba kuchaza ikamva, nowcasting iinzame ukuqikelela langoku (Choi and Varian 2012) . Ngamanye amazwi, nowcasting usebenzisa iindlela nokuqikelelwa iingxaki zemilinganiselo. Ngenxa yoko, kufuneka ube luncedo ngokukodwa oorhulumente abafuna amanyathelo kwangethuba nangokuchanekileyo ngamazwe abo. Nowcasting singayizekelisa ezininzi ngokucacileyo umzekelo Google Flu ekwizinga.

Cinga ukuba wena uziva kancinane phantsi imozulu ukuze uchwetheza "amayeza umkhuhlane" ibe injini yokukhangela, kufumana iphepha zamakhonkco impendulo, aze alandele omnye kubo iweb luncedo. Ngoku ocinga ukuba lo msebenzi sidlalwe phandle ngokwembono injini yokuphendla. Ngamaphanyazo onke, izigidi imibuzo ukufika evela ehlabathini lonke, yaye oku loqukuqelo imibuzo-oko Battelle (2006) iye ngokuba "database ezintle" - inikeza i window zihlaziywa rhoqo ezingqondweni collective lwehlabathi. Nangona kunjalo, Wabhekabheka umsinga kwenkcazelo ibe ngumlinganiso ukuxhaphaka ukuba umkhuhlane kunzima. Nje ngokubala inani imibuzo ngokuba "amayeza umkhuhlane" ukuze asebenze kakuhle. Asinguye wonke umntu kugqogqwa umkhuhlane ngenxa amayeza umkhuhlane hayi bonke abenza ngenxa amayeza umkhuhlane unalo umkhuhlane.

Iqhinga ezibalulekileyo krele emva Google Flu ekwizinga wayeza ukujika ingxaki womlinganiselo ibe yingxaki ku vhumbha. Kumaziko US for Disease Control and Prevention (CDC) inenkqubo esweni influenza eqokelela ulwazi oogqirha kwilizwe. Noko ke, ingxaki enye nale nkqubo CDC ngaba kukho ngeveki ezimbini ingxelo lag; ixesha ikuthatha data ukufika koogqirha ukucocwa, ukuqhubekekiswa, yavakalisa. Kodwa, xa uphethe ngobhubhani asakhasayo, iiofisi kwimpilo yoluntu andifuni ukwazi ukuba komkhuhlane kakhulu kwakukho kwiiveki ezimbini ezidlulileyo; bafuna ukwazi ukuba komkhuhlane yimalini na ekhoyo ngoku. Enyanisweni, kweminye imithombo emininzi yemveli data zentlalo, kukho izikhewu phakathi amaza yokuqokelelwa kwedatha kunye zabona ingxelo. Uninzi imithombo yedatha enkulu, kwelinye icala, zisoloko-on (Section 2.3.1.2).

Ngoko ke, uJeremy Ginsberg kunye noogxa (2009) wazama ukuqikelela CDC umkhuhlane data evela data yophendlo ye Google. Lo ngumzekelo "yokucingela langoku" kuba abaphandi ezama ukulinganisa ukuba umkhuhlane yimalini na ekhoyo ngoku ukuqikelela data elizayo evela CDC, iinkcukacha elizayo ukuba ukulinganisa langoku. Ukusebenzisa yokufunda umatshini, bona khangela zophendlo ezahlukeneyo nezigidi-50 ukuze ubone ukuba zeziphi ezizezona oqikelela CDC data umkhuhlane. Ekugqibeleni, bafumana iseti imibuzo 45 ezahlukeneyo ngathi kuba axelwe kwangaphambili, kunye iziphumo buntlana: bakwazi ukusebenzisa data uphendlo ukuqikelela data CDC. Ngokusekelwe ngokuyinxenye kweli phepha, leyo yapapashwa Nature, Google Flu ekwizinga waba ibali impumelelo soloko malunga amandla data ezinkulu.

Noko Kukho caveats ezimbini ezibalulekileyo kule impumelelo, nokuqonda ezi caveats kuya kukunceda ukuvavanya wenze nokuqikelelwa kunye nowcasting. Okokuqala, ukusebenza Google Flu ekwizinga waye bubhetele kakhulu kunokuba indlela elula ukuba uqikele- isixa umkhuhlane ngokusekelwe kwiisethi yomgama ukusuka imilinganiselo emibini eyona yakutshanje umkhuhlane ukuxhaphaka (Goel et al. 2010) . Ke, phezu athile ixesha Google Flu ekwizinga kwaba kubi eneneni ngaphezu le ndlela elula (Lazer et al. 2014) . Ngamanye amazwi, Google Flu ekwizinga idatha yaso yonke, ukufunda umatshini, becomputing enamandla abazange kakhulu ukuphumeza elula lula ukuyiqonda heuristic. Oku kubonisa ukuba xa kuhlolwa naluphi na uqikelelo okanye nowcast kubalulekile ukuthelekisa ngokuchasene lokwandlalela.

Lo icaveat yesibini ebalulekileyo Google Flu ekwizinga kukuba amandla ayo ukuqikelela CDC umkhuhlane data ndangenwa ukuba ukusilela elifutshane elide ukubola ixesha elide ngenxa isikhephe neyokuqungaqunga algorithmic. Ngokomzekelo, ebudeni 2009 Swine Flu ukuqhambuka Google Flu Trends ngokumangalisayo ekuhambeni-kuqikelelwa isixa influenza, mhlawumbi ngenxa yokuba abantu badla ukuguqula ihambo yabo uphendlo njengempendulo eyandileyo yokoyika yindyikityha jikelele (Cook et al. 2011; Olson et al. 2013) . Ukongeza kwezi ngxaki elifutshane, intsebenzo ngokuthe ukubola ekuhambeni kwexesha. Uxilongo izizathu le ukubola elide kunzima kuba algorithms yophendlo ye Google ke apho, kodwa kuyabonakala ukuba ngo-2011 Google wenza utshintsho Ndingacebisa zophendlo enxulumene xa abantu bafune iimpawu ezifana "umkhuhlane" kwaye "ukukhohlela" (kubonakala naye ukuba lo msebenzi akukho saba iyasebenza). Ukongeza kule osemqoka yinto elifanelekileyo ngokupheleleyo ukwenza ukuba uneshishini enjini yokukhangela, kwaye nefuthe adala uphendlo ezininzi ezinxulumene nempilo. Oku mhlawumbi ibe yimpumelelo ngenxa ishishini, kodwa yabangela Google Flu ekwizinga ukuya kwama-uqikelelo umkhuhlane ukuxhaphaka (Lazer et al. 2014) .

Ngethamsanqa, le ngxaki kunye Google Flu ntsingiselo fixable. Eneneni, usebenzisa iindlela ngononophelo, Lazer et al. (2014) kunye Yang, Santillana, and Kou (2015) baye bakwazi ukufumana iziphumo ezingcono. Ukuya phambili, ndiya silindele ukuba izifundo nowcasting oludibanisa data ezinkulu Umphandi eqokelelwe data-ezidibanisa Readymades Duchamp-isimbo kunye Michaelangelo-style Custommades-uya kwenza abenzi nkqubo ukuvelisa ngokukhawuleza kwaye echanileyo imilinganiselo langoku kunye neengqikelelo elizayo.