2.4.2 Forecasting a me ka nowcasting

Noi i ka wā e hiki mai ka paakiki, aka, luaÿi i ka makana, he hiki.

I ka lua o ka papa kuhikuhiE kou akamai hoʻohanaʻia e noiʻi me ka observational ikepili ua forecasting. Noi i ka wā e hiki mai ua notoriously paʻakikī, aka, he mea hiki e incredibly nui no ka olelo hooholo'ē aʻe, ina paha lakou e hana i ka poʻe a aupuni.

Kleinberg et al. (2015) e kaumaha ana elua moʻolelo e möakäka hou i ka nui o ka forecasting no kekahi kulekele pilikia. Hoi i noonoo ai i kekahi kulekele i hana, au e kapa Anna ia, ka mea i alo i ka wi, a e hooholo ina paha e hoolimalima i ka Hekikai no e hana i ua hula, e hoomahuahua i ka manawa o ka ua. Kekahi kulekele i hana, au e kapa ia ia Bob, e hooholo ina paha e lawe i umbrella e hana e hōʻalo ai i ka noho pulu ma ke alanui home. O Anna a me Bob ke hana i ka maikai hoʻoholo inā lakou hoomaopopo Weather, akā, pono e ike okoa mea. Anna pono e hoʻomaopopo paha i ka ua, hula kumu no ka ua. Bob, ma ka kekahi lima, aole ia e pono i ka mea ike i kekahi mea e pili ana i causality; oia wale nele i kekahi pala- pala kēla. Social noiʻi pinepine kālele ana ma luna o ka mea a Kleinberg et al. (2015) i kapa aku ai "ka ua hula-e like me" E pili pilikia, ka poe e kālele ana ma ka causality-a ignore "umbrella-e like me" E pili pilikia i hoʻomōhala i forecasting.

I e makemake ana e kālele ana, aka, ma ka hanaʻano o ka forecasting kapa nowcasting -a manawa loko mai o hoʻohuiʻia "Ano," a me "forecasting." Aka, mamua o ka luaÿi i ka wā e hiki mai, nowcasting a ho'āʻo paha e kilokilo i ka makana (Choi and Varian 2012) . Ma kekahi olelo, nowcasting hoʻohana forecasting epekema ka 'pilikia ana. E like me ia, he mea e e oi pono i ka aupuni, ka poe e noi mai ana 'eleu a me ka pololei ana a puni o ko lakou mau aina. Nowcasting hiki e ia manaʻo mea pono i ka hana o Google Flu kū i ke au.

Manao wale ia oe e manao ana he iki ma lalo iho o ka malie pela oe kikokiko i "flu Kids" iloko o ka search engine, loaa i kaʻaoʻao o nā loulou ma ka pane, a ma laila e hahai i kekahi o ia mau mea i ke kōkua webpage. Ano hoi i noonoo ai i keia ha awina ua pâʻani 'o ia i waho, mai ke kuanaʻike o ka search engine. Kela manawa, miliona o na'ōlelo āna i hiki mai a puni ke ao nei, a me keia kahawai o nā'ōlelo āna, i ka mea a Battelle (2006) Ua kapaia ka "hōkeoʻikepili o na manao" - 'oi aku ka mau hōʻano puka i loko o nā global ike paka i. Eia naʻe, haliu ae la ia i kēia kahawai o ka 'ike i loko o kekahi ana o ka prevalence o ka flu, ua pilikia. Aloha helu i ka nui o nā'ōlelo āna no ka mea, "flu Kids" ke ole ka hana pono. Aole ka mea ua loaÿa ka flu huli no flu laau, a aole hoi a pau i searchers no ka flu laau i ka flu.

Ka nui a me ka akamai liana a ma hope o Google Flu mea kūʻo ia e huli he ana pilikia iloko o ka forecasting pilikia. Ka US oaio? U no ka mea, mai Makuahine, a me Kāohi (CDC) he influenza ka mālamaʻana nenoaiu e 'ohi i ike mai kauka a puni ka aina. Eia naʻe, hoʻokahi pilikia i keia CDC nenoaiu ka mea, aia no he mau hebedoma hōʻike 'ana no lag; i ka manawa e lawe i ka 'ikepili hiki mai kauka e ka hoʻomaʻemaʻeʻana, hoʻokō', a paʻiʻia. Aka, i ka wa lawelawe ana i ka nuʻu laha, na ola? Anoa i makemake e ike i ka nui o influenza malaila, he elua pule aku nei. lakou i makemake ai e ike i ka nui o influenza he mea pono keia manawa. In mea, i na e ae kuʻuna kumu o ka lawelawe ike, he poe i wāhiʻia ma waena o nä hawewe o ka aeaiiuo ohi a me ka hoike mai i ka lags. Most nui ikepili kumu, ma kekahi lima, ua mau-ma (Pauku 2.3.1.2).

Nolaila, Ieremia Ginsberg a me nā hoapili (2009) hoao e kilokilo i ka CDC flu 'ikepili mai ka huli' ikepili Google. Keia mea he laʻana o "ka luaÿi i ka makana" no ka mea, o ka noiʻi, ua ho'āʻo ana e ana i ka nui o flu he mea ano ma ka luaÿi kēia mua aku ikepili mai ka CDC, kēia mua aku ikepili i ka mea ana i ka makana. E ho ohana i ke aʻo mīkini, mea, imi iho la ma 50 milionaʻokoʻa huli hua'ōlelo, e ike i ka poe predictive loa o ka CDC flu ikepili. Ultimately, loaa iho la ia lakou i ke kaʻina o 45ʻokoʻa'ōlelo āna i makemake ai e lilo loa predictive, a me nā hualoaʻa ua kokoke maikai; hiki ia lakou ke hoʻohana i ka huli ikepili e kilokilo i ka CDC aeaiiuo. Ma muli ma ka hapa ma keia pepa, ka mea i paʻiʻia ma loko o Nature, Google Flu mea kū i lilo i pinepine hai pomaikai moʻolelo e pili ana i ka mana o ka nui ikepili.

Elua nui caveats i keia maopopo ka pomaikai, nae, he mea ike i keia mau caveats e kōkua maiʻoe loiloi, ae hana forecasting, a me nowcasting. Mua, ka hana ana o Google Flu mea kū i maoli, aole loa mamua o ka mea kükohu e koho i ka huina o ka flu ma muli o ka laina extrapolation mai ka elua loa hala koke ana o ka flu prevalence (Goel et al. 2010) . A, ma luna o kekahi manawa paha Google Flu mea kū i ua maoliʻoi aku ma mua o kēia mea hoʻokokoke (Lazer et al. 2014) . Ma kekahi olelo, Google Flu LIKE me konaʻikepili a pau, mīkini aʻo, a me ka ikaika Me aole dramatically outperform i ka mea a me ka pono o ka hoomaopopo heuristic. Keia paipai i ka wa a ka loiloi 'ana i kekahi kēla a nowcast he mea nui e like aku i ka baseline.

I ka lua o ka nui caveat e pili ana Google Flu mea kū i ka mea a kona hiki pono ke kilokilo i ka CDC flu ikepili ua prone i pokole-makahiki hoʻokō 'ole' a me ka loihi-makahiki palaho no ka 'Aeʻa, a me algorithmic hoopalaimaka hoi. No ka mea, laʻana, i ka 2009 ua kumupuaa la Flu outbreak Google Flu mea kū i dramatically ma luna-aayoaeuiinoe i ke dala o ke influenza, paha no ka mea, na kanaka kamaʻilioʻana e hoololi i ko lakou huli hana i pane aku ai he ākea ka makaʻu i ka global pandemic (Cook et al. 2011; Olson et al. 2013) . Ma waho aʻeo kēia mau pokole-manawa pilikia, ke hana i nā amo ma luna o ka manawa. Diagnosing i nā kumu no kēia loihi makahiki palaho, ua paʻakikī no ka mea, o ka huli algorithms Google ua proprietary, aka, eia kahi mea akaka i ka 2011 Google i loli mea e paipai nā hua'ōlelo e huli ai ka wa kanakaʻimi i Ōuli e like me "kuni" a me "cough" (ia hoi he i keia kēia mau mea kanu, aole houʻeleu). Ohui i kēia mau mea kanu, he ihola kupono mea e hana ina oe e holo i ka huli engine hana, a ia ia i ke kanawai o ka hoʻomakaʻia aku ke ola Related searches. ʻo Ia nō paha he pomaikai no ka hana, aka, ua kena ae la Google Flu mea kū i luna-olelo flu prevalence (Lazer et al. 2014) .

Mea pōmaikaʻi, o keia mau pilikia me Google Flu mea kū i ua fixable. In mea, me ka nui akahele ano, Lazer et al. (2014) , a me Yang, Santillana, and Kou (2015) ua hiki ke loaʻa maikai hualoaʻa. E hele i mua, I manao nei nowcasting haʻawina e hui pu iho nui ikepili me ka noiʻi ohi 'ikepili-e hui pu iho Duchamp-mika Readymades me Michaelangelo-mika Custommades-, e hiki ai pilikino E pili'ē aʻe e paka wikiwiki a oi pololei ana o ka makana, a me wānana o kēia mua aku.