2.4.2 bonelang pele le nowcasting

Boleloa esale pele ka nako e tlang e thata, empa boleloa esale pele ka ea hona joale e be le bonolo.

Boemong ba bobeli ba ka sehloohong leqheka leo sebelisoa ke bafuputsi ka tlhahisoleseding observational e bonelang pele. Boleloa esale pele ka nako e tlang e tumme ka thata, empa ho ka ba Hoa makatsa hore ebe le habohlokoa hore baetsi fihlele qeto eo, ebang li sebetsa lik'hamphani tse kapa mebuso.

Kleinberg et al. (2015) e fana ka lipale tse peli tse hlakisa bohlokoa ba ho bonelang bakeng sa mathata a itseng leano. Ak'u nahane feela 'ngoe moetsi leano la, ke tla bitsa Anna hae, ea neng a tobane le komello le lokela ho etsa qeto ea hore na ba hira Shaman ho etsa motjeko pula ho eketsa monyetla oa hore a nese pula. moetsi e mong leano la, ke tla mo bitsa Bob, ba lokela ho etsa qeto ea hore na ba nka sekhele ho sebetsa ho qoba ho metsi tseleng lapeng. Ka bobeli Anna le Bob ka etsa qeto hamolemo haeba ba utloisisa boemo ba leholimo, empa ba lokela ho tseba lintho tse sa tšoaneng. Anna o lokela ho utloisisa hore na motjeko pula bakang pula. Bob, ka lehlakoreng le leng, ha ho hlokahale hore ba utloisise letho ka causality; a sa tsoa hloka e prognoza nepahetseng. Bafuputsi Social hangata ho ameha ka seo Kleinberg et al. (2015) bitsa "pula motjeko-joaloka" leano la mathata-le tseo ho ameha ka causality-'me ba hlokomoloha "sekhele-joaloka" mathata leano la hore a tsepame bonelang.

Ke kopa ho tsepamisa Leha ho le joalo, ka e mosa e khethehileng ea bonelang bitsoa nowcasting -e leng lentsoe lena le sebelisoa tsoa kopantseng "joale" le "bonelang." Ho e-na boleloa esale pele ka nakong e tlang, nowcasting boiteko ba ho bolela esale pele ea hona joale (Choi and Varian 2012) . Ka mantsoe a mang, nowcasting sebelisa mekhoa e bonelang bakeng mathata a lekanya. Ka lebaka leo, e lokela ba molemo haholo-holo ho mebuso ea ileng hloka mehato nakong le e nepahetseng ka linaheng tsa habo bona. Nowcasting ka bontšoa ka ho hlaka ka ho fetisisa le mohlala oa Google Ntaramane Trends.

Nka hore u ikutloe a batla tlas'a boemo ba leholimo hore o tle o thaepa "ntaramane litlhare" ka ho batla enjene, fumana leqephe le khokahano le a arabela, 'me joale latela e mong oa bona ho ea sebakeng se webpage thusa. Joale nka mosebetsi ona ntse a bapala ka ntle ho tswa lehlakoreng la ea search engine. Motsotso mongwe le mongwe, batho ba limilione dipotso tse fihla tsoang lefatšeng lohle, 'me molatsoana o ena ea dipotso-seo Battelle (2006) e bitsoang "polokelongtshedimosetso oa boikemisetso" - fana ka fensetere kamehla ntjhafatswa ho kena ka har'a kopanetsweng ilibana lefatše lohle. Leha ho le joalo, ho retelehela molatsoana ena ea tsebiso ka tekanyo ea ho ata ha ntaramane ke ho le thata. A mpa a bala fihlang palo ea dipotso for "litlhare ntaramane" a se ke a sebetsa hantle. Hase bohle ba nang le phenyang ntaramane bakeng sa litlhare ntaramane 'me hase bohle ba batlang bakeng sa litlhare ntaramane na ntaramane.

Tsa bohlokoa le bohlale qhekella ka morao Google Ntaramane Trends ne e le ho retelehela bothata lekanya ho kena bothata bonelang. Litsi Tsa Taolo US le Thibelo Mafu (CDC) e na le ntaramane tsamaiso go baya leitlho hore bokella boitsebiso bo tsoang lingaka ho potoloha le naha. Leha ho le joalo, bothata e 'ngoe le tsamaiso ena CDC e na le ka beke tse peli tse itlalehang du lieu; nako ea ho nka bakeng sa data ya fihla ho tswa lingaka a lokela ho hloekisoa, sebetswa, mme e hatisitsoeng. Empa, ha ba sebetsana seoa hlahella, liofisi tsa bophelo bo botle tsa phatlalatsa ha ba batle ho tseba ka bokae ntaramane ho ne ho setse libeke tse peli tse fetileng; ba batla ho tseba hore na lintho tse ngata ntaramane ho na le hona joale. Ha e le hantle, linaheng tse ling tsa mehloli e mengata e tloaelehileng ea data ho tsoang sechabeng, ho na le likheo pakeng tsa maqhubu a pokello data le ho tlaleha lags. Most khōlō mehloling ya data, ka lehlakoreng le leng, kamehla-on (Karolo 2.3.1.2).

Ka lebaka leo, Jeremy Ginsberg le basebetsi-'moho (2009) ba ile ba leka ho noha CDC ntaramane data ho tswa ya data ho batla Google. Sena ke mohlala oa "boleloa esale pele ka ea hona joale" hobane bafuputsi ba ne ba leka ho lekanya hore na lintho tse ngata ntaramane ho na le hona joale ke boleloa esale pele ka ya data nakong e tlang ho tloha CDC, ya data nakong e tlang e lekanyang tse teng. U sebelisa ithuta mochine, ba sheba ka ya ka dipehelo limilione tse 50 tse fapaneng ho batla ho bona hore tse fetisisa lepa ya data ya CDC ntaramane. Qetellong, ba ile ba fumana sete ya 45 dipotso tse fapaneng hore ba bonahala e le ho fetisisa ka ho lepa, 'me liphello e bile tse molemo haholo: ba ne ba ka sebelisa ya data ho batla ho noha ya data CDC. Thehiloeng ka karolo e itseng pampiring ena, e neng e hatisoa ka Nature, Google Ntaramane Trends ile ea e-hangata pheta lebisang katlehong pale ea matla a ya data e khōlō.

Ho na le caveats tse peli tsa bohlokoa ho atleha sena se bonahalang eka ho le joalo, le ho utloisisa caveats tse tla u thusa ho hlahloba le ho etsa bonelang le nowcasting. Ntlha ea pele, tshebetso ya Google Ntaramane Trends e le hantle ha ho molemo ho feta ho ka tsela e itekanetseng ea mohlala eo hakanya palo ea ntaramane thehiloe extrapolation guttate ho tloha tse pedi litekanyo tsa moraorao tsa ata ntaramane (Goel et al. 2010) . 'Me, lilemong tse ka bang linako nako Google Ntaramane Trends e le hantle ho feta bona katamelo bonolo hakana (Lazer et al. 2014) . Ka mantsoe a mang, Google Ntaramane Trends ka tlhahisoleseding tsohle tsa eona, ho ithuta mochine, 'me dikhomphiutha e matla ha aa ka tsela e hlollang outperform o bonolo le o bonolo haholoanyane hore ba utloisise heuristic. Sena se bontša hore ha ho hlahloba prognoza leha e le efe kapa nowcast ke habohlokoa ho bapisa khahlanong le ya motheo e.

Tsa bohlokoa caveat ea bobeli e mabapi le Google Ntaramane Trends ke hore matla a ho noha CDC ntaramane ya data ne le tsekamelo ea ho hloleha nako e khutsoanyane le nako e telele ho bola ka lebaka la hoholeha 'me algorithmic tsietsoa. Ka mohlala, nakong ea 2009 likolobe Ntaramane qhoma Google Ntaramane Trends ka tsela e hlollang lilemo tse fetang-hakanngoa tjhelete ya ntaramane, mohlomong hobane batho ba le tšekamelo ea ho fetola fuputso boitšoaro ba bona e le karabelo ea tšabo atile la seoa ea lefatše lohle (Cook et al. 2011; Olson et al. 2013) . Ntle ho mathata ana a 'short-lentsoe lena, e le tiragatso ee butle-butle ha nako e ntse decayed. Diagnosing mabaka lentsoe lena bola e telele ho leng thata hobane dikgatotharabololong fuputso Google ba dikhampani, empa ho bonahala eka ka 2011 Google etsa liphetoho e neng e tla fana ka maikutlo a dipehelo tse amanang fuputso ha batho batla matšoao a tšoana le "feberu" le "khohlela" (e boetse e bonahala hore tšobotsi ena ha e sa le mafolofolo boboleling). Phaella ka ho re tšobotsi ena ke ntho e feletseng le kahlolo e molemo ho etsa haeba u matha ya kgwebo batla enjene, 'me e ne phello ea generating bophelo bo botle amanang di fetlekago more. Ena e ne e ka etsahala hore ebe a atleha ho namola bakeng sa khoebo, empa e bakoa Google Ntaramane Trends ho ata holimo-khakanyo ntaramane (Lazer et al. 2014) .

Ka lehlohonolo, mathata ana le Google Ntaramane Trends ke fixable. Ha e le hantle, ba sebelisa mekhoa e hlokolosi haholoanyane, Lazer et al. (2014) le ho Yang, Santillana, and Kou (2015) ba ile ba khona ho fumana diphetho molemo. Tswela pele, ke lebella hore lithuto tsa nowcasting hore kopana ya data khōlō le mofuputsi bokella ya data-hore kopana Duchamp mokhoa Readymades le Michaelangelo mokhoa Custommades-ho tla etsa hore baetsi ba le leano la ho hlahisa ka lebelo le nepahetseng haholoanyane litekanyo tsa hona joale le tse boletsoeng esale pele ka bokamoso.