Elon Musk<\/a> platformu yeni sat\u0131n alm\u0131\u015ft\u0131 ve onu daha \u015feffaf hale getirmek i\u00e7in \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 savunuyordu. O d\u00f6nemde de algoritmalar a\u00e7\u0131klanm\u0131\u015ft\u0131 ama uzmanlar a\u00e7\u0131klanan bilgilerin yetersiz oldu\u011funu ve kodlar\u0131n neden o \u015fekilde \u00e7al\u0131\u015ft\u0131\u011f\u0131na dair fazla bilgi vermedi\u011fini s\u00f6yl\u00fcyordu.<\/p>\nO tarihten bu yana X algoritmalar\u0131 muhtemelen bir\u00e7ok kez g\u00fcncellendi ama bu g\u00fcncellemeler halka a\u00e7\u0131klanmad\u0131.<\/p>\n
\u00a0<\/p>\n
\u015eimdi ise Musk ge\u00e7en hafta bu y\u00f6nde verdi\u011fi s\u00f6z\u00fc tutarak sitenin algoritmas\u0131n\u0131 tekrar a\u00e7\u0131k kaynakl\u0131 hale getirdi. Musk’\u0131n ge\u00e7en hafta, “Kullan\u0131c\u0131lara hangi organik g\u00f6nderilerin ve reklam i\u00e7eriklerinin \u00f6nerilece\u011fini belirlemek i\u00e7in kullan\u0131lan t\u00fcm kodlar dahil olmak \u00fczere yeni X algoritmas\u0131n\u0131 7 g\u00fcn i\u00e7inde a\u00e7\u0131k kaynakl\u0131 hale getirece\u011fiz” demi\u015fti.<\/p>\n
\u00a0<\/p>\n
X, sal\u0131 g\u00fcn\u00fc GitHub’da yay\u0131nlad\u0131\u011f\u0131 bir g\u00f6nderide, i\u00e7erik olu\u015fturma kodunun anla\u015f\u0131l\u0131r bir a\u00e7\u0131klamas\u0131n\u0131 ve program\u0131n nas\u0131l \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 g\u00f6steren bir diyagram\u0131 payla\u015ft\u0131.<\/p>\n
\u00a0<\/p>\n
Musk ayr\u0131ca, gelecekte her d\u00f6rt haftada bir algoritma de\u011fi\u015fikliklerini de a\u00e7\u0131klayaca\u011f\u0131n\u0131 dile getiriyor.<\/p>\n
Peki a\u00e7\u0131klanan kodlara g\u00f6re X algoritmas\u0131 nas\u0131l i\u015fliyor ve g\u00f6nderilerin daha \u00e7ok etkile\u015fim almas\u0131 i\u00e7in ne yapmal\u0131?<\/p>\n
\u00a0<\/p>\n
4 ANA MOD\u00dcL<\/strong><\/p>\n\u00a0<\/p>\n
Github\u2019da a\u00e7\u0131klanan bilgilere g\u00f6re algoritma, platformun karma\u015f\u0131k yap\u0131s\u0131n\u0131 y\u00f6netmek i\u00e7in 4 ana mod\u00fcle ayr\u0131ld\u0131. Bunlar \u015fu \u015fekilde:<\/p>\n
\u00a0<\/p>\n
Home Mixer: <\/strong>Bu katman orkestra \u015fefi rol\u00fcnde. T\u00fcm par\u00e7alar\u0131 bir araya getirerek ak\u0131\u015f\u0131 son haline getiriyor.<\/p>\n\u00a0<\/p>\n
Thunder: <\/strong>Takip etti\u011finiz ki\u015filerden gelen i\u00e7erikleri (buna \u201cIn-Network\u201d ad\u0131 verilmi\u015f) ger\u00e7ek zamanl\u0131 i\u015fleyen sistem. Son derece h\u0131zl\u0131 \u00e7al\u0131\u015f\u0131yor.<\/p>\n\u00a0<\/p>\n
Phoenix: <\/strong>Bu katman ise takip etmedi\u011finiz ki\u015fileri ke\u015ffetmenizi sa\u011flayan (\u201cOut-of-Network\u201d ad\u0131 verilmi\u015f) ve \u00f6n\u00fcn\u00fcze d\u00fc\u015fen i\u00e7erikleri puanlayan k\u0131s\u0131m. xAI’\u0131n Grok tabanl\u0131 Transformer modelini kullan\u0131yor.<\/p>\n\u00a0<\/p>\n
Candidate Pipeline:<\/strong> \u0130\u00e7erik adaylar\u0131n\u0131 se\u00e7en, filtreleyen ve s\u0131ralayan standart bir \u00e7er\u00e7eve (framework) sunuyor.<\/p>\n\u00a0<\/p>\n
X AKI\u015eINIZ NASIL OLU\u015eTURULUYOR?<\/strong><\/p>\n\u00a0<\/p>\n
Bir kullan\u0131c\u0131 X’i a\u00e7t\u0131\u011f\u0131nda algoritma belirli a\u015famalardan ge\u00e7iyor.<\/p>\n
\u00a0<\/p>\n
\u0130lk ad\u0131m aday belirleme; bu sistem iki koldan i\u00e7erik topluyor. \u201cA\u011f i\u00e7i\u201d i\u00e7erik toplama s\u00fcrecinde takip etti\u011finiz ki\u015filerin en son ve g\u00fcndeminizle en alakal\u0131 tweet’lerini getiriyor. \u201cA\u011f d\u0131\u015f\u0131\u201d i\u00e7erik s\u00fcrecinde ise takip etmedi\u011finiz ancak ilginizi \u00e7ekebilecek tweet’leri buluyor.<\/p>\n
\u00a0<\/p>\n
Sistem bunu yaparken “embedding” (vekt\u00f6rel benzerlik) y\u00f6ntemini kullan\u0131yor; yani ge\u00e7mi\u015fte be\u011fendi\u011finiz i\u00e7eriklere benzer i\u00e7erikleri k\u00fcresel bir havuzdan se\u00e7iyor.<\/p>\n
\u00a0<\/p>\n
Bu ad\u0131m\u0131n ard\u0131nan filtreleme s\u00fcreci geliyor. \u00d6nce “\u00e7\u00f6p” say\u0131labilecek i\u00e7erikler eleniyor. Bunlar aras\u0131nda engelledi\u011finiz veya sessize ald\u0131\u011f\u0131n\u0131z hesaplar, daha \u00f6nce g\u00f6rd\u00fc\u011f\u00fcn\u00fcz tweet’ler, kendi att\u0131\u011f\u0131n\u0131z tweet’ler ve \u015fiddet, spam veya hassas olarak i\u015faretlenmi\u015f i\u00e7erikler yer al\u0131yor.<\/p>\n
\u00a0<\/p>\n
Bundan sonra gelen en \u00f6nemli k\u0131s\u0131m ise puanlama (scoring). Bu a\u015famada Grok tabanl\u0131 Phoenix modeli her bir tweet i\u00e7in bir dizi olas\u0131l\u0131k hesapl\u0131yor. Hesaplama i\u00e7in algoritma \u015fu sorular\u0131n cevab\u0131n\u0131 ar\u0131yor:<\/p>\n
\u00a0<\/p>\n
– Bu kullan\u0131c\u0131 bu tweet’i be\u011fenecek mi?<\/p>\n
– Yan\u0131t verecek mi?<\/p>\n
– Retweet yapacak m\u0131*?<\/p>\n
– Tweet’e t\u0131klayacak m\u0131?<\/p>\n
– Tweet’i okumak i\u00e7in zaman harcayacak m\u0131 (buna \u201cdwell time\u201d deniyor)?<\/p>\n
\u00a0<\/p>\n
Ak\u0131\u015f\u0131n belirlenmesinde X, art\u0131k “el yap\u0131m\u0131” \u00f6zellikler yerine tamamen kullan\u0131c\u0131n\u0131n ge\u00e7mi\u015f etkile\u015fim zincirine bakan derin \u00f6\u011frenme modelleri kullan\u0131yor.<\/p>\n
\u00a0<\/p>\n
ALGOR\u0130TMA NEY\u0130 SEV\u0130YOR?<\/strong><\/p>\n\u00a0<\/p>\n
Bu s\u00fcrecin ard\u0131ndan da \u00f6n\u00fcn\u00fcze d\u00fc\u015fecek g\u00f6nderilerin a\u011f\u0131rl\u0131kland\u0131rmas\u0131 ve s\u0131ralanmas\u0131 geliyor. Zira her bir etkile\u015fim t\u00fcr\u00fcn\u00fcn bir “a\u011f\u0131rl\u0131\u011f\u0131” var. \u00d6rne\u011fin, bir tweet’i be\u011fenmek 1 puan ise, bir videoyu sonuna kadar izlemek 5 puan olabilir. Negatif eylemler (engelleme, sessize alma, “ilgimi \u00e7ekmiyor” deme) ise puan\u0131 ciddi \u015fekilde d\u00fc\u015f\u00fcr\u00fcyor. T\u00fcm bu puanlar toplan\u0131yor ve en y\u00fcksek puanl\u0131 tweet’ler en \u00fcste yerle\u015fiyor.<\/p>\n
\u00a0<\/p>\n
Kodlardaki \u201cWeighted Scorer\u201d (a\u011f\u0131rl\u0131k puan\u0131) mant\u0131\u011f\u0131na g\u00f6re en y\u00fcksek puan getiren eylemler \u015funlar:<\/p>\n
\u00a0<\/p>\n
Yeniden payla\u015f\u0131m: <\/strong>En y\u00fcksek puan\u0131 \u201cretweet\u201d getiriyor. Bir i\u00e7eri\u011fi kendi kitlenize ta\u015f\u0131mak, algoritma i\u00e7in en b\u00fcy\u00fck kalite sinyali. Bir Retweet, bir bir be\u011feniden yakla\u015f\u0131k 20 ila 30 kat daha de\u011ferli.<\/p>\n\u00a0<\/p>\n
Dwell Time (Kalma s\u00fcresi): <\/strong>Kritik. Kullan\u0131c\u0131n\u0131n tweet \u00fczerinde ge\u00e7irdi\u011fi s\u00fcrenin (okuma veya video izleme) platformda tutma g\u00fcc\u00fc y\u00fcksek. Bu y\u00fczden okurun 2 dakikadan fazla s\u00fcre g\u00f6nderinizde kalmas\u0131, tweet’e devasa bir puan veriyor.<\/p>\n\u00a0<\/p>\n
Yan\u0131t:<\/strong> Y\u00fcksek. \u00d6zellikle tweet sahibi yan\u0131ta geri cevap verirse puan katlanarak art\u0131yor. Ancak tek tarafl\u0131 ve k\u0131sa yan\u0131tlar\u0131n puan\u0131 daha d\u00fc\u015f\u00fck.<\/p>\n\u00a0<\/p>\n
Be\u011feni: <\/strong>Orta. En yayg\u0131n ama \u201cucuz\u201d etkile\u015fim. Algoritma bunu temel bir onaylama sinyali olarak g\u00f6r\u00fcyor; ancak tek ba\u015f\u0131na bir tweet’i viral yapmaya yetmiyor.<\/p>\n\u00a0<\/p>\n
Profile t\u0131klama: <\/strong>D\u00fc\u015f\u00fck\/Orta. Kullan\u0131c\u0131n\u0131n sizin profilinize de t\u0131klamas\u0131 merak\u0131n\u0131 g\u00f6steriyor. E\u011fer profil ziyareti takiple sonu\u00e7lan\u0131rsa puan \u00e7ok art\u0131yor, sadece t\u0131klama di\u011ferlerine g\u00f6re daha zay\u0131f olsa da yine de \u00f6nemli bir sinyal.<\/p>\n\u00a0<\/p>\n
G\u0130ZL\u0130 FAKT\u00d6RLER<\/strong><\/p>\n\u00a0<\/p>\n
G\u00f6nderilerin ak\u0131\u015fta g\u00f6r\u00fcnme olas\u0131l\u0131\u011f\u0131nda \u00e7ok \u00f6nemli bir di\u011fer fakt\u00f6r de \u201cnegatif a\u011f\u0131rl\u0131klar\u201d.<\/p>\n
Hatta asl\u0131nda s\u0131ralaman\u0131n en ba\u015f\u0131nda asl\u0131nda pozitifler de\u011fil, negatifler var. Kodlara g\u00f6re bir “\u015fikayet” veya “engelleme” alman\u0131z, kazand\u0131\u011f\u0131n\u0131z y\u00fczlerce be\u011feni puan\u0131n\u0131 tek seferde silebiliyor.<\/p>\n
\u00a0<\/p>\n
\u015eikayet edilmeniz yakla\u015f\u0131k -369 puan k\u0131rarken, takibin b\u0131rak\u0131lmas\u0131 i\u00e7eri\u011fin da\u011f\u0131t\u0131m\u0131n\u0131 durdurabilecek noktaya getirebiliyor. Ayn\u0131 zamanda kullan\u0131c\u0131lar\u0131n sizi \u201cilgimi \u00e7ekmiyor\u201d diye bildirmesi de a\u011f\u0131r bir negatif puan demek.<\/p>\n
\u00a0<\/p>\n
Sonu\u00e7ta, e\u011fer g\u00f6r\u00fcnt\u00fclenme say\u0131n\u0131z\u0131 art\u0131rmak istiyorsan\u0131z, stratejinizi “yeniden payla\u015f\u0131m” ve “okunma s\u00fcresi\u201d \u00fczerine kurman\u0131z gerekebilir. Bu y\u00fczden uzun bilgiseller, ilgi \u00e7ekici videolar veya g\u00f6rseller, basit bir metin tweetinden daha avantajl\u0131. Ay\u0131 zamanda g\u00f6nderilerinize takip\u00e7ilerinizin fikrini de sorman\u0131z iyi olabilir.<\/p>\n
\u00a0<\/p>\n<\/div>\n
\n
<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"Sosyal medya platformu X (eski ad\u0131yla Twitter), uygulamadaki ak\u0131\u015f\u0131 d\u00fczenleyen ve kullan\u0131c\u0131lar\u0131n g\u00f6nderilerinin ald\u0131\u011f\u0131 etkile\u015fimde b\u00fcy\u00fck rol oynayan algoritma kodlar\u0131n\u0131 2023\u2019ten bu yana ilk kez halka a\u00e7t\u0131. \u00a0 2023\u2019te Tesla CEO\u2019su Elon Musk platformu yeni sat\u0131n alm\u0131\u015ft\u0131 ve onu daha \u015feffaf hale getirmek i\u00e7in \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 savunuyordu. O d\u00f6nemde de algoritmalar a\u00e7\u0131klanm\u0131\u015ft\u0131 ama uzmanlar a\u00e7\u0131klanan bilgilerin […]<\/p>\n","protected":false},"author":908,"featured_media":37011,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[12],"tags":[],"class_list":["post-37010","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji"],"jetpack_publicize_connections":[],"jetpack_featured_media_url":"https:\/\/kibrisayna.com\/wp-content\/uploads\/2026\/01\/WhatsAppImage2026-01-21at21.37.38-586240.jpeg","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/posts\/37010","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/users\/908"}],"replies":[{"embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/comments?post=37010"}],"version-history":[{"count":0,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/posts\/37010\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/media\/37011"}],"wp:attachment":[{"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/media?parent=37010"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/categories?post=37010"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/tags?post=37010"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}