2.4.1.3 nnyocha nke mmadụ mgbasa ozi site Chinese ọchịchị

Researchers kpụchasịrị Chinese elekọta mmadụ media saịtị na-amụ nnyocha. Ha si mesoo incompleteness na latent-àgwà inference.

Na mgbakwunye na nnukwu data eji na abụọ tupu atụ, na-eme nnyocha nwekwara ike na-anakọta ha onwe ha observational data, dị ka e ebube gosiri Gary King, Jennifer Pan, na Michael Roberts ' (2013) nnyocha na nnyocha site Chinese ọchịchị.

Social media posts na China na-enyochakwa site buru ibu ala ngwa na-eche na-agụnye ọtụtụ iri puku ndị mmadụ. Researchers na ụmụ amaala, Otú ọ dị, nwere obere echiche nke otú ndị a censors ikpebi ihe ọdịnaya a ga-ehichapụ si elekọta mmadụ media. Ọkà mmụta nke China n'ezie nwere dị iche iche ndị na-atụ anya banyere nke iche iche nke posts yikarịrị ka ehichapụ. Ụfọdụ na-eche na censors-elekwasị anya na posts ndị oké egwu nke ala ebe ndị ọzọ chere na ha elekwasị anya na posts na-agba ume mkpokọta omume, dị ka ngagharị iwe. Ịmụta otú m si nke ndị a na-atụ anya bụ eziokwu nwere pụtara n'ihe otú na-eme nnyocha na-aghọta China na ndị ọzọ na authoritarian ọchịchị na-ekere òkè nnyocha. Ya mere, Eze na ndị ọrụ ibe chọrọ tụnyere posts na e bipụtara na ekemende ehichapụ na-posts na e bipụtara, na-adịghị na-ehichapụ.

Ịnakọta ndị a posts aka ịtụnanya engineering feat nke crawling ihe karịrị 1,000 Chinese mmadụ mgbasa ozi weebụsaịtị-kwa dị iche iche na peeji nke layouts-achọta mkpa posts, na mgbe ahụ akafiak ekese ndị a posts ịhụ nke e mechara ehichapụ. Na mgbakwunye na nkịtị engineering nsogbu metụtara na nnukwu ọnụ ọgụgụ web-crawling, nke a oru ngo nwere kwukwara ihe ịma aka nke na ọ dị mkpa ịkpachara nnọọ ngwa ngwa n'ihi na ọtụtụ enyochakwa posts na-ebutu na-erughị 24 awa. Ndị ọzọ okwu, a ngwa ngwa crawler ga na-atụ uche nza nke posts a na-enyochakwa. Ọzọkwa, crawlers nwere ime ihe nile a data collection mgbe evading nchọpụta aghara ime ka mmadụ mgbasa ozi weebụsaịtị igbochi ohere ma ọ bụ n'ụzọ na-agbanwe ha atumatu omume na-amụ ihe.

Ozugbo a oke engineering ọrụ e dechara, King na ndị ọrụ ibe ya nwetara banyere 11 nde posts on 85 dị iche iche isiokwu ya na ha tupu kpọmkwem dabeere na ha tụrụ anya na na ọkwa nke uche. Dị ka ihe atụ, a isiokwu nke elu uche bụ Ai Weiwei, ndị dissident artist; a isiokwu nke etiti uche bụ ekele na mweda nke Chinese ego, na a isiokwu nke ala uche bụ World Cup. N'ime ndị a 11 nde posts banyere 2 nde ama e censored, ma posts on ukwuu enwe mmetụta ọsọ ọsọ isiokwu e enyochakwa ubé karịa posts on n'etiti na ala uche isiokwu. Yabụ, Chinese censors bụ banyere ihe a ga-onye nnyocha a post na kwuru Ai Weiwei dị ka a post na-ekwu banyere World Cup. Nchọpụta ndị a na-dakọtara na simplistic echiche na ndị ọchịchị censors niile posts on mwute isiokwu.

Nke a dị mfe ngụkọta oge nke nnyocha ọnụego site isiokwu nwere ike ịbụ na-eduhie eduhie, Otú ọ dị. Dị ka ihe atụ, ndị ọchịchị nwere ike onye nnyocha posts na-akwado Ai Weiwei, ma na-ahapụ posts na-akatọ ndị ya. Iji ịmata ọdịiche dị n'etiti posts ihe ọma, na-eme nnyocha mkpa tụọ mmetụta nke ọ bụla post. N'ihi ya, otu ụzọ anyị si eche echiche banyere ya bụ na mmetụta nke onye ọ bụla post na ihe dị mkpa latent feature nke ọ bụla post. O di nwute na, n'agbanyeghị ọtụtụ ọrụ, n'ụzọ zuru ezu akpaghị aka na usoro nke mmetụta nchọpụta iji tupu ẹdude Akwụkwọ ọkọwa okwu bụ ka na-adịghị nnọọ mma ọtụtụ mgbe, (chee echiche laa azụ nsogbu na-eke ihe n'ụzọ mmetụta uche usoro iheomume nke September 11, 2001 site Nkebi 2.3.2.6). Ya mere, Eze na ndị ọrụ ibe mkpa a ụzọ ịkpọ ha 11 nde mmadụ mgbasa ozi posts iji mara ma ha bụ ndị 1) oké egwu nke ala, 2) na-akwado nke ala, ma ọ bụ 3) adịghị mkpa ma ọ bụ eziokwu akụkọ banyere ihe ndị mere. Nke a na-ada ka a oke ọrụ, ma, ha na-edozi ya na-eji a dị ike atọ; onye nke jupụtara data sayensị ma ugbu a dịtụ obere na-elekọta mmadụ na sayensị.

Akpa, na a nzọụkwụ a na-akpọ tupu nhazi, ndị na-eme nnyocha converted na-elekọta mmadụ media posts n'ime a akwụkwọ-okwu matriks, bụ ebe e nwere otu n'usoro nke ọ bụla akwụkwọ na otu kọlụm nke ahụ e dekọrọ ma post dị otu okwu (eg, mkpesa, okporo ụzọ, wdg). Next, otu ìgwè ndị nnyocha na-enyere aka aka-labeled na mmetụta nke a sample nke post. Mgbe ahụ, Eze na ndị ọrụ ibe ji mee nke a-aka-kpọrọ data na-eme atụmatụ a igwe mmụta nlereanya na ike infer na mmetụta nke a post dabeere na e ji mara ya. N'ikpeazụ, ha ji nke a igwe mmụta nlereanya na-eme atụmatụ na mmetụta nke ihe nile nde 11 posts. N'ihi ya, kama iji aka ịgụ na labeling 11 nde posts (nke ga-logistically agaghị ekwe omume), ha manually kpọrọ mmadụ ole na ole posts na mgbe ahụ na-eji ihe data ọkà mmụta sayensị na-akpọ chịkwara mmụta na-eme atụmatụ na ige niile posts. Mgbe-emecha nke a analysis, King na ndị ọrụ ibe nwee ike ikpebi na, dịtụ ijuanya na ndị gbasara nke puru omume a post na-ehichapụ bụ nweghị mmekọ ma ọ bụ oké egwu nke obodo ma ọ bụ inye nkwado nke obodo.

Ọgụgụ 2.3: Simplified schematic maka usoro eji na King, Pan, and Roberts (2013) na-estimating na mmetụta nke 11 nde Chinese mmadụ mgbasa ozi posts. Akpa, na a nzọụkwụ a na-akpọ tupu nhazi, ndị na-eme nnyocha converted na-elekọta mmadụ media posts n'ime a akwụkwọ-okwu matriks (lee Grimmer and Stewart (2013) maka ozi ndị ọzọ). Nke abụọ, ndị nnyocha ahụ aka-coded na mmetụta nke a obere sample nke posts. Nke atọ, na-eme nnyocha nyere ọzụzụ a chịkwara mmụta nlereanya ịgụnye ndị mmetụta nke posts. Nke anọ, ndị na-eme nnyocha na-eji chịkwara mmụta nlereanya na-eme atụmatụ na mmetụta nke niile posts. Lee King, Pan, and Roberts (2013) , Odide Ntụkwasị B maka a ọzọ zuru ezu nkọwa.

N'ikpeazụ, Eze na ndị ọrụ ibe chọpụtara na ọ bụ nanị atọ ụdị posts e mgbe nile enyochakwa:-akpali agụụ mmekọahụ, na-akatọ censors, na ndi nwere mkpokọta edinam nwere (ie,-ekwe omume nke na-eduga na nnukwu-ọnụ ọgụgụ ngagharị iwe). Site n'idebe ihe a nnukwu ọnụ ọgụgụ posts na-ehichapụ na posts a na-ehichapụ, King na ndị ọrụ ibe nwee ike ịmata otú censors na-arụ ọrụ dị nnọọ site na-ekiri ma na-agụta. Ụdi nnyocha, na ha n'ezie na-etinye aka n'ime Chinese na-elekọta mmadụ media ilekota site na-eke posts na n'usoro dị iche iche ọdịnaya na e ji atụ nke na-enyochakwa (King, Pan, and Roberts 2014) . Anyị ga-amụkwu banyere ibuo-eru nso n'Isi 4. Ọzọkwa, na-ese onyinyo a isiokwu ndị ga-eme n'oge nile nke akwụkwọ, ndị a latent-àgwà inference nsogbu-nke nwere ike mgbe ụfọdụ a ga-edozi na chịkwara mmụta-atụgharị si na-nnọọ nkịtị na-elekọta mmadụ nnyocha na digital afọ. Ị ga-ahụ foto yiri nnọọ chepụta ma gụọ 2.3 ke Ibuot 3 (Ịjụ ajụjụ) na 5 (Ịmepụta uka mmekota); ọ bụ otu n'ime ndị ole na ole echiche na-egosi na multiple isi.

All atọ nke ihe atụ ndị a na-arụ ọrụ omume nke taxi ọkwọ ụgbọala na New York, enyi guzobere site ụmụ akwụkwọ, na-elekọta mmadụ media nnyocha omume nke Chinese ọchịchị-egosi na dịtụ mfe agụta nke observational data nwere ike na-eme nnyocha iji nwalee usoro iwu amụma. Mgbe ụfọdụ, nnukwu data-enyere gị aka ime nke a agụta dịtụ ozugbo (dị ka na ihe banyere New York Tagzi). N'ọnọdụ ndị ọzọ, na-eme nnyocha ga-mkpa na-anakọta ha onwe ha observational data (dị ka na ihe banyere Chinese nnyocha); obibi incompleteness site merging data ọnụ (dị ka na ihe banyere netwọk evolushọn); ma ọ bụ na-arụ ụdị ụfọdụ nke latent-àgwà inference (dị ka na ihe banyere Chinese nnyocha). Dị ka m nwere olileanya na ihe atụ ndị a na-egosi, n'ihi na-eme nnyocha bụ ndị nwere ike ịjụ akpali ajụjụ, nnukwu esetịpụ oké nkwa.