{"id":53,"date":"2020-09-03T16:38:52","date_gmt":"2020-09-03T13:38:52","guid":{"rendered":"http:\/\/acikkitap.com.tr\/oaek\/chapter\/bolum-9-dogal-dil-isleme-ve-ogrenme-analitigi\/"},"modified":"2020-09-03T16:38:52","modified_gmt":"2020-09-03T13:38:52","slug":"bolum-9-dogal-dil-isleme-ve-ogrenme-analitigi","status":"publish","type":"chapter","link":"https:\/\/acikkitap.com.tr\/oaek\/chapter\/bolum-9-dogal-dil-isleme-ve-ogrenme-analitigi\/","title":{"raw":"B\u00f6l\u00fcm 9 Do\u011fal Dil \u0130\u015fleme ve \u00d6\u011frenme Analiti\u011fi","rendered":"B\u00f6l\u00fcm 9 Do\u011fal Dil \u0130\u015fleme ve \u00d6\u011frenme Analiti\u011fi"},"content":{"raw":"\n

Carolyn Penstein Rose<\/span><\/span><\/p>\nDil Teknolojileri Enstit\u00fcs\u00fc ve \u0130nsan-Bilgisayar Etkile\u015fimi Enstit\u00fcs\u00fc, Carnegie Mellon \u00dcniversitesi, ABD<\/span><\/span>\n\nDOI: 10.18608\/hla17.009<\/span><\/span>\n

\u00d6Z<\/h2>\nBu b\u00f6l\u00fcm s\u00f6ylem analitiklerini (SA) tan\u0131tmaktad\u0131r. S\u00f6ylem analitikleri, ara\u015ft\u0131rmay\u0131 destekleyecek analitik mercekler sunmak, bi\u00e7imlendirici ve de\u011fer bi\u00e7meye y\u00f6nelik de\u011ferlendirmeleri i\u015fler k\u0131lmak, \u00f6\u011frenme etkinliklerinin etkilili\u011fini artt\u0131rmak amac\u0131yla yap\u0131lan m\u00fcdahaleleri dinamik ve ba\u011flama duyarl\u0131 bir \u015fekilde harekete ge\u00e7irmek ve \u00f6\u011frenme etkinliklerinden sonraki raporlar ve geri bildirimler gibi yans\u0131tma ara\u00e7lar\u0131n\u0131n hem \u00f6\u011frenmeyi hem de \u00f6\u011fretimi destekleyecek \u015fekilde temin edilmesi gibi bir\u00e7ok alana etki etmektedir. Bu b\u00f6l\u00fcm\u00fcn amac\u0131, bu alanda ne yap\u0131labilece\u011fine dair belirli bir miktar \u00fcmidi ve ku\u015fkuculu\u011fu y\u00fcreklendirmek ayn\u0131 zamanda okuyucuya anlaml\u0131 bir i\u015f yapabilmek i\u00e7in ihtiya\u00e7 duyulan derinlikte uzmanl\u0131\u011f\u0131 sunmakt\u0131r. Amac\u0131 gerekli uzmanl\u0131\u011f\u0131 vermek de\u011fildir. Aksine, burada ama\u00e7 okuyucunun yeterli derinli\u011fi sa\u011flayacak bir ekip olu\u015fturmak i\u00e7in ne t\u00fcr i\u015fbirlikli \u00e7al\u0131\u015fma ortaklar\u0131 arayaca\u011f\u0131n\u0131 kavramada kendi yerini belirlemesidir. Alan\u0131n bir tan\u0131mlamas\u0131yla ba\u015flay\u0131p hem teorik hem de metodolojik olarak geni\u015f bir alan\u0131 ku\u015fat\u0131p hem temsil hem algoritmik boyutlar\u0131n\u0131 ara\u015ft\u0131racak ve daha derinlere dalmaya niyetli okuyuculara sonraki ad\u0131m \u00f6nerileriyle bitirece\u011fiz.<\/span>\n\nAnahtar kelimeler<\/span>: S\u00f6ylem analizi, i\u015fbirlikli \u00f6\u011frenme, makine \u00f6\u011frenmesi, analiz ara\u00e7lar\u0131<\/span>\n

S\u00f6ylem analiti\u011fi (SA) \u00f6\u011frenme analitikleri i\u00e7inde bir aland\u0131r (\u00d6A; Buckingham Shum, 2013; Buckingham Shum, de Laat, de Liddo, Ferguson ve Whitelock, 2014). E\u011fitsel ortamlar i\u00e7indeki a\u00e7\u0131k u\u00e7lu sorular\u0131n i\u015flenmesini kapsar ve alandaki ara\u015ft\u0131rmalar b\u00fcy\u00fck oranda yaz\u0131 \u00e7al\u0131\u015fmalar\u0131n\u0131n de\u011ferlendirilmesi konusuna odaklanm\u0131\u015ft\u0131r ancak bundan daha fazlas\u0131n\u0131; tart\u0131\u015fma forumlar\u0131nda, sohbet odalar\u0131nda, mikro bloglarda, bloglarda ve hatta wikilerde yap\u0131lan tart\u0131\u015fmalar\u0131 da kapsar. \u00d6A\u2019y\u0131 genel olarak, \u00f6\u011frenenlerin \u00f6\u011frenmelerini dinleme yoluyla \u00f6\u011frenmeyi \u00f6\u011frenme olarak ele al\u0131r, dinleyi\u015fimiz ise \u00e7o\u011funlukla veri madencili\u011fi ve makine \u00f6\u011frenmesi teknolojileri ile desteklenir ancak alanda yay\u0131nlanm\u0131\u015f \u00e7al\u0131\u015fmalar \u00f6nc\u00fc olsa da t\u00fcm durumlar bir otomasyon s\u00f6z konusu de\u011fildir (Knight ve Littleton, 2015; Milligan, 2015). Ayr\u0131ca, biz bu alan\u0131 farkl\u0131 k\u0131lan \u015feyin, verinin \u00fcretildi\u011fi t\u00fcm ak\u0131\u015flar\u0131nda, dinlemeye ait olan do\u011fal dil verisine odaklan\u0131lmas\u0131 oldu\u011funu d\u00fc\u015f\u00fcnmekteyiz.<\/p>\n

Bu b\u00f6l\u00fcm \u00d6A sahas\u0131 i\u00e7inde konumlanm\u0131\u015f olan bu alana \u00e7ok temel bir ba\u015flang\u0131\u00e7 niteli\u011findedir. SA de\u011fi\u015fimli olarak iki tehlikeli kavram yan\u0131lg\u0131s\u0131ndan muzdariptir. Birincisi asl\u0131nda bir\u00e7ok ki\u015fi i\u00e7in, kullan\u0131ma haz\u0131r olan ve analiz i\u015fini bir d\u00fc\u011fmeye basarak onlar ad\u0131na yapacak bir \u00e7\u00f6z\u00fcme sahip olma arzusuyla k\u00f6r\u00fcklenen a\u015f\u0131r\u0131 ve u\u00e7 bir beklentidir. Bu kavram yan\u0131lg\u0131s\u0131n\u0131n tutsa\u011f\u0131 olanlar hayal k\u0131r\u0131kl\u0131\u011f\u0131na u\u011framaya mahk\u00fbmdur. En basit veya en g\u00fc\u00e7l\u00fc modelleme teknolojilerinin etkili kullan\u0131m\u0131 \u00e7ok\u00e7a haz\u0131rl\u0131k, emek ve uzmanl\u0131k gerektirmektedir. \u0130kinci kavram yan\u0131lg\u0131s\u0131 ise bazen birinci kavram yan\u0131lg\u0131s\u0131ndan do\u011fan hayal k\u0131r\u0131kl\u0131klar\u0131n\u0131 veya s\u00f6ylemin karma\u015f\u0131kl\u0131klar\u0131n\u0131 son derece derinlikli bir \u015fekilde bilmekten kaynaklanan, hi\u00e7bir bilgisayar\u0131n var olan ince ayr\u0131nt\u0131lar\u0131 tamamen yakalayamayaca\u011f\u0131 fikrini yok sayman\u0131n zorlu\u011fu sonucu olarak ortaya \u00e7\u0131km\u0131\u015f olan a\u015f\u0131r\u0131 bir ku\u015fkuculuktur. S\u00f6ylem inan\u0131lmaz bir \u015fekilde karma\u015f\u0131k olsa da teknoloji harikas\u0131 modelleme yakla\u015f\u0131mlar\u0131n\u0131n tan\u0131mlayabildi\u011fi anlaml\u0131 \u00f6r\u00fcnt\u00fcler oldu\u011fu da bir ger\u00e7ektir. Bu b\u00f6l\u00fcm boyunca teknolojinin bug\u00fcnk\u00fc durumunu a\u00e7\u0131klayan \u00d6\u011frenme Analiti\u011fi, Bilgi ve di\u011fer ba\u011flant\u0131l\u0131 konferanslardan al\u0131nan bir\u00e7ok yay\u0131mlanm\u0131\u015f \u00e7al\u0131\u015fmaya at\u0131fta bulunulmu\u015ftur. Bilgi i\u015flemsel sosyo dilbilime dair yap\u0131lan yak\u0131n zamanl\u0131 bir ara\u015ft\u0131rma, hik\u00e2yeyi dil teknolojileri dal\u0131n\u0131n perspektifinden anlatmaktad\u0131r (Nguyen, Dogru\u00f6z, Ros\u00e9 ve de Jong, bas\u0131m a\u015famas\u0131nda) ve konuya \u00f6zel ilgi duyan okuyucular\u0131n ilgisini \u00e7ekecektir.<\/p>\n

Bu b\u00f6l\u00fcm, biraz daha derine dalmay\u0131 isteyen okuyuculara faydal\u0131 ipu\u00e7lar\u0131 sunmay\u0131 umut etmektedir. SA konusu ile ilgili ge\u00e7mi\u015f iki at\u00f6lye \u00e7al\u0131\u015fmas\u0131 \u00d6A toplulu\u011fu i\u00e7indeki temel \u00e7al\u0131\u015fmalar\u0131 ara\u015ft\u0131rm\u0131\u015ft\u0131r (Buckingham Shum, 2013; Buckingham Shum vd., 2014). Daha daralt\u0131lm\u0131\u015f kapsamda Bilgisayar destekli i\u015fbirlikli \u00f6\u011frenme ile ilgili konu ve y\u00f6ntemleri i\u00e7eren kapsaml\u0131 bir genel de\u011ferlendirme daha \u00f6nce hakemli dergilerde bas\u0131lm\u0131\u015f \u00fc\u00e7 makalede bulunabilir (Ros\u00e9 vd., 2008; Mu, Stegman, Mayfield, Ros\u00e9 ve Fischer, 2012; Gweon, Jain, McDonough, Raj ve Ros\u00e9, 2013). Alana dair k\u0131sa bir ders ise, edX platformunda verilen 2014 Veri Analiti\u011fi ve KA\u00c7D'yi \u00d6\u011frenme1<\/sup><\/a> deki metin madencili\u011fi b\u00f6l\u00fcm\u00fcnde bulunabilir. Di\u011fer kaynaklar b\u00f6l\u00fcm\u00fcn sonunda sunulacakt\u0131r.<\/p>\n

Bu b\u00f6l\u00fcmde; \u00f6\u011frenme olaylar\u0131 s\u0131ras\u0131nda ifade edilen do\u011fal dille ilgileniyoruz. Kuramsal ve metodolojik olarak kapsay\u0131c\u0131 olmak istiyoruz. Teorik ve metodolojik olarak kapsay\u0131c\u0131 olmay\u0131 arzu ediyoruz. S\u00f6ylem analitikleri ile ilgili yap\u0131lm\u0131\u015f h\u00e2lihaz\u0131rdaki \u00e7al\u0131\u015fmalar\u0131n \u00e7o\u011fu, \u00f6\u011frenmeyi ve onun dil ile ba\u011flant\u0131s\u0131n\u0131 bili\u015fsel bir mercekten g\u00f6rmektedir, ba\u015fka bir deyi\u015fle, s\u00f6ylem i\u00e7erisinde var olan dil davran\u0131\u015f\u0131 kategorileri aray\u0131\u015f\u0131; ortak s\u00f6ylem s\u00fcre\u00e7leri ve \u00f6\u011frenme ile ili\u015fkili olan bili\u015fsel s\u00fcre\u00e7ler aras\u0131ndaki ba\u011flant\u0131dan dolay\u0131 \u00f6\u011frenme kazan\u0131mlar\u0131 ile ilgili baz\u0131 \u00f6ng\u00f6r\u00fclerde bulunmaktad\u0131r. Bu b\u00f6l\u00fcmde \u00f6\u011frenme ve onun dille olan ba\u011flant\u0131s\u0131n\u0131, \u00f6\u011frenmede rol\u00fc olan bili\u015fsel ve sosyal fakt\u00f6rler aras\u0131ndaki \u00f6nemli etkile\u015fimi g\u00fc\u00e7lendirmek ad\u0131na sosyal bir mercek arac\u0131l\u0131\u011f\u0131 ile g\u00f6rmeyi ama\u00e7l\u0131yoruz. (Hmelo-Silver, Chinn, Chan ve O\u2019Donnell, 2013; O\u2019Donnell ve King, 1999). \u00d6rne\u011fin, \u00f6\u011frenme etkile\u015fimlerinde \u00f6nemli bir destekleyici rol oynayan temel e\u011filimler, tutumlar ve ili\u015fkileri ortaya \u00e7\u0131karacak s\u00f6ylem s\u00fcre\u00e7lerini belirlemeyi ama\u00e7l\u0131yoruz. Hangi durumda ifade edilmi\u015f olursa olsun do\u011fal dil son derece ki\u015fisel ve k\u00fclt\u00fcreldir. \u0130\u00e7erisine ki\u015fisel deneyimlerimizin ve bizden \u00f6nceki ku\u015faklar\u0131n yap\u0131tlar\u0131 yerle\u015fmi\u015f durumdad\u0131r. Dil se\u00e7imlerimizdeki detaylar bilin\u00e7li olarak yans\u0131tt\u0131\u011f\u0131m\u0131z, ayn\u0131 zamanda bilin\u00e7li olarak saklad\u0131\u011f\u0131m\u0131z ve hatta fark\u0131nda bile olmad\u0131\u011f\u0131m\u0131z kimliklerimiz hakk\u0131nda ipu\u00e7lar\u0131 verir. Hedef kitlemize dair ve onlara y\u00f6nelik tutumumuz ve hedef kitlemize g\u00f6re kendimizi konumland\u0131r\u0131\u015f\u0131m\u0131z hakk\u0131ndaki varsay\u0131mlar\u0131m\u0131z\u0131 veya bazen sadece hedef kitlemizin bizim yapt\u0131\u011f\u0131m\u0131z\u0131 d\u00fc\u015f\u00fcnmelerini istedi\u011fimiz varsay\u0131mlar\u0131 yans\u0131t\u0131rlar. Biz bu se\u00e7imleri ili\u015fkiler ekonomisi i\u00e7erisinde, benimsedi\u011fimiz hedeflere ula\u015fmak i\u00e7in bir para birimi gibi kullan\u0131r\u0131z (Ribeiro, 2006).<\/p>\n

Bu anlay\u0131\u015fla hesaplamay\u0131, \u00f6\u011frenenleri dinlemeyi desteklemek i\u00e7in bir mercek olarak kullan\u0131rken, bizimle \u00f6\u011frenme s\u00fcre\u00e7leri aras\u0131nda duran teknolojilere her daim; bir t\u00fcr dijital bi\u00e7ime kay\u0131t yap\u0131l\u0131rken neyin kayboldu\u011fu ve neyin d\u00f6n\u00fc\u015ft\u00fc\u011f\u00fc, hatta analitik teknolojinin uygulanmas\u0131 s\u0131ras\u0131nda ger\u00e7ekle\u015fen sonraki indirgeme ve d\u00f6n\u00fc\u015f\u00fcm de d\u00e2hil olmak \u00fczere yorumlama konusundaki sorumlulu\u011fumuzun bir k\u0131sm\u0131n\u0131; b\u0131rak\u0131yor oldu\u011fumuzu kabul etmemiz gerekir (Morrow ve Brown, 1994). Bu \u015ferhi de d\u00fc\u015ferek, bu b\u00f6l\u00fcmde yo\u011fun bir \u015fekilde model yorumlama ve ge\u00e7erlik de\u011ferlendirmesine dair sorulara odaklanaca\u011f\u0131z.<\/p>\n\n

BU B\u00d6L\u00dcM\u00dcN KAPSAMI VE ODA\u011eI<\/h2>\n

Herhangi bir ki\u015fi analitikleri d\u00fc\u015f\u00fcnd\u00fc\u011f\u00fcnde, zihninde aniden algoritmalar belirir (Witten, Frank ve Hall, 2011). Ancak uygulamal\u0131 istatistikten ders \u00e7\u0131karmak ve onun yerine \u00f6nce g\u00f6sterimler hakk\u0131nda d\u00fc\u015f\u00fcnmek \u00f6nem arz eder. SA \u00e7al\u0131\u015fmas\u0131n\u0131n kalbinde verinin g\u00f6sterimine odaklanmak yatar. Makine \u00f6\u011frenmesi modelleri do\u011frudan metinlere uygulanamaz. Daha ziyade, metinden kestirim \u00f6zellikleri elde edilmelidir. Bu \u00f6ng\u00f6r\u00fcc\u00fc \u00f6zellikler, sorular olarak alg\u0131lanabilir: \u201cMetinde_var m\u0131?\u201d veya \u201cMetinde_ka\u00e7 kez bulunuyor?\u201d Her bir \u00f6zellik bu sorulardan biriyse, her durumda, \u00f6zellik de\u011feri, sorunun cevab\u0131d\u0131r. \u0130lgilenen okuyucular; kamuya a\u00e7\u0131k LightSIDE arac\u01312<\/sup><\/a> (Mayfield ve Ros\u00e9, 2013; Gianfortoni, Adamson ve Ros\u00e9, 2011), gibi geni\u015f bir kullan\u0131c\u0131 k\u0131lavuzu, \u00f6rnek veri k\u00fcmeleri, s\u00fcre\u00e7le ilgili y\u00f6nergeler ve yard\u0131m etmeye haz\u0131r ara\u015ft\u0131rmac\u0131lar\u0131n ileti\u015fim bilgilerini i\u00e7eren; \u00fccretsiz olarak eri\u015filebilen, kullan\u0131ma haz\u0131r bir tezg\u00e2h (workbench) ile deneyler yaparak metinden elde edilebilecek basit \u00f6zelliklerin derinli\u011fine ve s\u0131n\u0131flama modellerinin yordama kesinli\u011fi \u00fczerinde nas\u0131l bir etkisi oldu\u011funa dair iyi bir fikir edinebilirler.<\/p>\n

Metne uygulanan modelleme tekniklerinde ba\u015far\u0131n\u0131n anahtar\u0131 anlaml\u0131 ipu\u00e7lar\u0131 \u00fcretecek do\u011fru sorular\u0131 sormakt\u0131r. Bu soruya ili\u015fkin d\u00fc\u015f\u00fcnme dilin nas\u0131l yap\u0131land\u0131r\u0131ld\u0131\u011f\u0131n\u0131 dikkate almakla ba\u015flar. Y\u00fczeysel olarak \u00e7\u0131plak g\u00f6ze dil yekpare, yap\u0131land\u0131r\u0131lmam\u0131\u015f bir b\u00fct\u00fcn olarak g\u00f6r\u00fcnse de asl\u0131nda \u00e7oklu katmanlardan olu\u015fan, her biri dilbilimin ayr\u0131 bir alan\u0131n\u0131n i\u00e7inde tan\u0131mlanan bir yap\u0131dan olu\u015fur. Bir dilbilim ders kitab\u0131n\u0131n (O\u2019Grady, Archibald, Aronoff, & Rees \u2013 Miller, 2009) giri\u015f niteli\u011findeki bir incelemesi, bu \u00d6A alan\u0131na girmek isteyen ara\u015ft\u0131rmac\u0131lar i\u00e7in de\u011ferli bir kaynak olacakt\u0131r. En k\u00fc\u00e7\u00fck par\u00e7ac\u0131\u011f\u0131 ses yap\u0131s\u0131 d\u00fczeyindedir, fonoloji (ses bilim) olarak bilinmektedir. Burada dilin temel ses b\u00f6l\u00fcmleri ve bunlar\u0131n dilin hece yap\u0131s\u0131na nas\u0131l uyum sa\u011flad\u0131\u011f\u0131 tan\u0131mlanmaktad\u0131r. Seslerin temel bir alfabesi bir dizi ses birimlerini olu\u015fturur fakat leh\u00e7eler i\u00e7inde bunlar belirli \u015fekillerde telaffuz edilebilirler ve bu etnisite, sosyoekonomik d\u00fczey ve b\u00f6lge gibi sosyal anlaml\u0131l\u0131kla ilgili olan de\u011fi\u015fkenler k\u00fcmesi ile olan ba\u011flant\u0131s\u0131ndan dolay\u0131 sosyal anlamda bir \u00f6nem ta\u015f\u0131maktad\u0131r. Bu d\u00fczeyin hem en \u00fcst\u00fcnde, morfoloji (bi\u00e7im bilim) olarak bilinen kelimelerin daha i\u00e7 yap\u0131s\u0131n\u0131n tan\u0131mland\u0131\u011f\u0131 bir katman vard\u0131r. Buras\u0131 dil bilgisi derslerinde \u00f6\u011frendi\u011fimiz ve bu arada fillerin zamanlar\u0131n\u0131 veya isimlerin say\u0131lar\u0131n\u0131 de\u011fi\u015ftiren tak\u0131lar\u0131n ortaya \u00e7\u0131kt\u0131\u011f\u0131 yerdir. Yukar\u0131da, t\u00fcm c\u00fcmlelerin gramer yap\u0131s\u0131n\u0131n tan\u0131mland\u0131\u011f\u0131 s\u00f6zdizimi d\u00fczeyidir. Ayn\u0131 zamanda c\u00fcmle d\u00fczeyinde anlam\u0131n de\u011fi\u015fmez ifadeler yoluyla, kurallar taraf\u0131ndan ve s\u00f6z dizim kurallar\u0131 taraf\u0131ndan y\u00f6nlendirilen daha k\u00fc\u00e7\u00fck b\u00f6l\u00fcmler d\u00fczenlenerek olu\u015fturuldu\u011fu anlam bilim alan\u0131 vard\u0131r ve s\u00f6zc\u00fcksel anlam bilim d\u00fczeyinde alt d\u00fczey anlam bilim b\u00f6l\u00fcmleriyle ili\u015fkilidir. C\u00fcmle seviyesinin \u00fcst\u00fcnde, yap\u0131n\u0131n di\u011fer y\u00f6nleri aras\u0131nda retorik stratejileri buldu\u011fumuz s\u00f6ylem d\u00fczeyidir. Bu teknik terimler bir\u00e7ok okuyucuya yabanc\u0131 gelebilir ancak daha ileri okumalar i\u00e7in uygun kaynaklar\u0131 bulmak isteyen okuyuculara yararl\u0131 arama terimleri sunacaklard\u0131r.<\/p>\n

Do\u011fal dil verisinin otomatikle\u015ftirilmi\u015f analizin hedefi oldu\u011fu bir\u00e7ok alan\u0131n tarihini izlersek, ge\u00e7erli modelleme i\u00e7in kilit unsurun anlaml\u0131 g\u00f6sterimler tasarlamak olarak adland\u0131r\u0131ld\u0131\u011f\u0131, ayn\u0131 nakarat\u0131 duyar\u0131z. Bu \u00f6rne\u011fi bu b\u00f6l\u00fcme koymakta, okuyucular\u0131n ayn\u0131 dersi zorlu bir \u015fekilde \u00f6\u011frenmekten korunmalar\u0131 \u00fcmidi yatmaktad\u0131r. SA ilgili olan bu dersin iyi \u00f6\u011frenildi\u011fi en eski durumlardan biri otomatikle\u015ftirilmi\u015f kompozisyon yaz\u0131s\u0131 puanlama ile ilgiliydi (Page, 1966; Shermis ve Hammer, 2012). En eski yakla\u015f\u0131mlar regresyon gibi basit modelleri ve ortalama c\u00fcmle uzunlu\u011fu, uzun kelime say\u0131s\u0131n\u0131 ve kompozisyonun uzunlu\u011funu sayma gibi basit \u00f6zellikleri kulland\u0131lar. Bu yakla\u015f\u0131mlar, say\u0131sal puanlar\u0131n atanmas\u0131n\u0131n g\u00fcvenilirli\u011fi a\u00e7\u0131s\u0131ndan olduk\u00e7a ba\u015far\u0131l\u0131 olmu\u015ftur (Shermis ve Burstein, 2013); Ancak de\u011ferlendirme i\u00e7in kan\u0131t kullan\u0131m\u0131nda ge\u00e7erlili\u011fi olmad\u0131\u011f\u0131 i\u00e7in ele\u015ftirildiler. Sonraki \u00e7al\u0131\u015fmalarda, odak noktas\u0131 daha \u00e7ok \u00f6\u011fretenlerin yazmay\u0131 puanlad\u0131klar\u0131 kendi r\u00fcbriklerine (dereceli puanlama anahtar\u0131) neleri d\u00e2hil ettikleri gibi \u00f6zelliklerin belirlenmesine kayd\u0131. Bu inceleme genellikle metni harf harf b\u00f6len (unigram) dil g\u00f6sterimlerini temel ald\u0131klar\u0131 i\u00e7in unigram \u00f6zelliklerle ilgili problemlerin kurban\u0131 olsalar da yine de i\u00e7erik tabanl\u0131 de\u011ferlendirmeleri desteklemek amac\u0131yla fakt\u00f6r analizine benzeyen \u00f6rt\u00fck semantik analiz (latent semantic analysis) (\u00d6SA: Foltz, 1996) veya gizli Drichlet tahsisi (latent Drichlet allocation) (GDT; Blei, Ng ve Jordan, 2003; Griffiths ve Steyvers, 2004) gibi teknikleri de kapsayarak i\u00e7erik odakl\u0131 \u00f6zelliklerin d\u00e2hil edilmesine neden oldu. CohMeTrix (McNamara ve Graesser, 2012) gibi di\u011fer fakt\u00f6r analitik dil analizi yakla\u015f\u0131mlar\u0131, bili\u015fsel g\u00fc\u00e7l\u00fck gibi fakt\u00f6rler d\u00e2hil \u00e7e\u015fitli boyutlar\u0131n yan\u0131nda, \u00f6\u011frencilerin yaz\u0131lar\u0131n\u0131n de\u011ferlendirilmesinde kullan\u0131lm\u0131\u015ft\u0131r. Belli bir d\u00fczeyde s\u00f6z dizimsel yap\u0131sal analizleri kullanan son derece g\u00fcndelik \u00e7al\u0131\u015fma alanlar\u0131nda CohMetrix yararlar sa\u011flam\u0131\u015ft\u0131r(Ros\u00e9 ve VanLehn, 2005). Fen e\u011fitiminde a\u00e7\u0131k u\u00e7lu sorular\u0131n de\u011ferlendirilmesinde LightSIDE ile ba\u015far\u0131 elde edilmi\u015ftir (Nehm, Ha ve Mayfield, 2012; Mayfield ve Ros\u00e9, 2013).<\/p>\n

Bu noktada SA\u2019ya dair a\u015f\u0131r\u0131 ve d\u00fc\u015f\u00fck beklentiler aras\u0131ndaki gerilime geri d\u00f6nmek faydal\u0131 olacakt\u0131r. Uygun ve anlaml\u0131 \u00f6zelliklerin belirlenmesindeki zorluklar\u0131 d\u00fc\u015f\u00fcnd\u00fc\u011f\u00fcm\u00fczde modelleme ara\u00e7lar\u0131 arac\u0131l\u0131\u011f\u0131 ile olu\u015fturdu\u011fumuz merceklerin s\u0131n\u0131rl\u0131l\u0131klar\u0131 ile uzla\u015fmam\u0131z gereklidir. SA ara\u015ft\u0131rmac\u0131lar veya uygulay\u0131c\u0131lar\u0131n ellerinde, kendileri ile d\u0131\u015f d\u00fcnyada ger\u00e7ekle\u015fen \u00f6\u011frenme par\u00e7alar\u0131 aras\u0131nda duran bir mercek olarak hizmet verir ya da \u00f6\u011frenen ve \u00f6\u011fretenler, \u00f6\u011frenenler, ya da \u00f6\u011frenen ile \u00f6\u011frenme teknolojileri aras\u0131nda bir filtre olabilirler. Mercekler kendileri arac\u0131l\u0131\u011f\u0131 ile g\u00f6r\u00fclen d\u00fcnyan\u0131n b\u00fct\u00fcn ayr\u0131nt\u0131lar\u0131n\u0131 basit\u00e7e aktarmad\u0131klar\u0131 i\u00e7in kesinlikle faydal\u0131d\u0131rlar. Aksine g\u00f6r\u00fcnt\u00fclerin onlar olmadan etkili bir \u015fekilde g\u00f6r\u00fclemeyecek \u00f6zelliklerini vurgularlar. Bu da onlar\u0131n yapmas\u0131na ihtiya\u00e7 duydu\u011fumuz \u015feydir. Bunlar\u0131 yapmak i\u00e7in ihtiyac\u0131m\u0131z olan \u015fey budur. Ayn\u0131 zamanda, tasar\u0131m taraf\u0131ndan daha az ilgin\u00e7 olarak kabul edilen \u00f6zellikleri ise karart\u0131rlar. Mercekler her zaman e\u011fip b\u00fckerler. Fakat onlar\u0131 ge\u00e7erli bir \u015fekilde kullanmak i\u00e7in, uygun bir mercek se\u00e7ebilmek ad\u0131na her birinin neyi vurgulad\u0131\u011f\u0131n\u0131 ya da neyi karartt\u0131\u011f\u0131n\u0131 ve g\u00f6rd\u00fc\u011f\u00fcm\u00fcz \u015feyi ge\u00e7erli bir \u015fekilde yorumlayabilmemiz i\u00e7in resmin onsuz veya ba\u015fka bir mercekle nas\u0131l olabilece\u011fini her zaman sorgulayarak anlamam\u0131z \u015fartt\u0131r. Bu y\u00fczden biz, en ba\u015f\u0131ndan itibaren bu alandaki ara\u015ft\u0131rmay\u0131 kullananlar, bu mercekleri geli\u015ftirenler veya onlar\u0131 etkin olarak ara\u015ft\u0131rmada veya uygulamada kullananlar\u0131 uygulama s\u0131ras\u0131nda neyin ka\u00e7\u0131n\u0131lmaz olarak kayboldu\u011fu ya da d\u00f6n\u00fc\u015ft\u00fc\u011f\u00fc konusunda kontroll\u00fc olmalar\u0131 konusunda uyar\u0131yoruz. Bu b\u00f6l\u00fcm dikkatini SA'n\u0131n kapsam\u0131nda olan \u00f6zel alanlara y\u00f6neltecektir.<\/p>\n\n

MET\u0130N G\u00d6STER\u0130M\u0130<\/h2>\n

Verinin analitik mercekler arac\u0131l\u0131\u011f\u0131 ile nas\u0131l g\u00f6r\u00fcnece\u011fini fazlas\u0131yla etkileyecek \u00f6nemli kararlar temsil a\u015famas\u0131nda al\u0131n\u0131r. Bu a\u015famada, metin yekpare g\u00f6r\u00fcnen bir b\u00fct\u00fcnden onun i\u00e7inden \u00e7\u0131kar\u0131labilecek bir dizi belirleyici niteli\u011fe d\u00f6n\u00fc\u015ft\u00fcr\u00fcl\u00fcr. Her \u00f6zellik \u00e7\u0131kar\u0131c\u0131, metne bir soru ve metnin verdi\u011fi cevap ise g\u00f6sterimde ona tekab\u00fcl eden niteli\u011fin de\u011feridir. Bir insan hakk\u0131nda tek bildi\u011finiz \u015feyin yirmi soruluk bir oyunda sorulan sorulara verilen bir dizi cevap oldu\u011funu hayal edin ve \u015fimdi g\u00f6reviniz bu insan\u0131 bir tak\u0131m sosyal g\u00f6r\u00fc\u015f kategorileri i\u00e7inde s\u0131n\u0131fland\u0131rmak olsun. E\u011fer sorular dikkatlice yap\u0131land\u0131r\u0131lm\u0131\u015fsa, hatas\u0131z bir kestirimde bulunabilirsiniz ama yine de bu ki\u015fiye ait bir\u00e7ok bilgi ve \u00f6ng\u00f6r\u00fcn\u00fcn s\u00fcre\u00e7 i\u00e7erisinde kaybolaca\u011f\u0131n\u0131 da kabul etmeniz gerekir. Bu \u00f6nemli a\u015famada bilgi bir kez kayboldu\u011funda hik\u00e2ye, dil teknolojileri alan\u0131n\u0131n bak\u0131\u015f a\u00e7\u0131s\u0131yla ne kadar ileri d\u00fczey olursa olsun bir algoritman\u0131n uygulanmas\u0131 ile yeniden elde edilemez. Bu nedenle, bu b\u00f6l\u00fcm boyunca temsil ve g\u00f6sterimlerle ilgili kararlar\u0131n dikkatlice al\u0131nmas\u0131n\u0131n, yorum \u00fczerinde dikkatlice ve etrafl\u0131ca d\u00fc\u015f\u00fcnmenin ve yapt\u0131\u011f\u0131m\u0131z \u00e7\u0131kar\u0131mlar\u0131n ge\u00e7erlili\u011fini dikkatlice sorgulaman\u0131n \u00f6nemini vurguluyoruz. Bu alanda yeni olan okuyucular bu ikazlar\u0131 hayali bulabilirlerse de deneyimle bu daha netle\u015feceklerdir.<\/p>\n\n

Genel Bak\u0131\u015f<\/h3>\n

Metni harf harf b\u00f6lme (unigram) \u00f6zellikleri metin madencili\u011fi problemlerinde kullan\u0131lan en tipik \u00f6zellik \u00e7\u0131kar\u0131c\u0131lar\u0131d\u0131r. Bir unigram \u00f6zellik aral\u0131\u011f\u0131nda e\u011fitim verilerindeki bir dizi metnin i\u00e7inde g\u00f6r\u00fclen her kelime i\u00e7in bu kelimenin metnin i\u00e7indeki varl\u0131\u011f\u0131n\u0131 soracak uygun bir \u00f6zellik olacakt\u0131r. Unigram \u00f6zellik aral\u0131klar\u0131 genellikle makul bir \u015fekilde y\u00fcksek performans elde ederken, modeller e\u011fitim verilerininkine \u00e7ok benzer ko\u015fullar alt\u0131nda toplanan verilerin \u00f6tesine genelleme yapmakta ba\u015far\u0131s\u0131zd\u0131rlar. Genellemedeki yetersizli\u011fin nedeni bu unigram modellerinin temel olarak her s\u0131n\u0131f de\u011fer etiketini y\u00fczeysel bir \u015fekilde ezberlemesidir. Bu modeller insanlar\u0131n belirli bir dizi durumda nelerden bahsettiklerini e\u011fitim verilerinde bulunan bu etiketle ili\u015fkilendirirler. E\u011fer bunda bir tutarl\u0131l\u0131k varsa, bu o zaman modeller taraf\u0131ndan \u00f6\u011frenilebilir ancak bu tutarl\u0131l\u0131k nadiren daha \u00f6teye ge\u00e7ip genelle\u015fir. \u00d6zellikler ortaya \u00e7\u0131kar\u0131ld\u0131\u011f\u0131nda olu\u015fan genelleme, amaca uygun bir yap\u0131 katman\u0131ndan gelir.<\/p>\n

Metnin \u00f6zellik tabanl\u0131 g\u00f6steriminin amac\u0131, m\u00fcmk\u00fcn olan en y\u00fcksek do\u011frulukla kestirimci modellemeyi ger\u00e7ekle\u015ftirmek hedefiyle, s\u0131kl\u0131kla s\u0131n\u0131fland\u0131rma veya say\u0131sal de\u011ferlendirme amac\u0131yla kestirimci modellemeyi etkinle\u015ftirmektir (Ros\u00e9 vd., 2008; Mc-Laren vd., 2007; Allen, Snow, McNamera, 2015). Bu b\u00f6l\u00fcm\u00fcn oda\u011f\u0131 ise uyumland\u0131rma s\u00fcreci olacakt\u0131r. Ancak SA'n\u0131n geni\u015f alan\u0131 i\u00e7inde bulunan baz\u0131 \u00e7al\u0131\u015fmalarda, temsil ve g\u00f6sterimler odak noktas\u0131d\u0131r, anlam belirlenen \u00f6ng\u00f6r\u00fcsel \u00f6zellikten \u00e7\u0131kar\u0131l\u0131r ve kestirimsel\/ \u00f6ng\u00f6r\u00fcsel modellemenin, e\u011fer varsa, belirlenen \u00f6zelliklerin bir do\u011frulamas\u0131 olarak hizmet etti\u011finin dikkate al\u0131nmas\u0131 \u00f6nemlidir (Simsek, Sandor ve Buckingham Shum, 2015; Dascalu, Dessus, McNamera, 2015; Snow, Allen, Jacovina, Perret, McNamera, 2015).<\/p>\n

S\u0131n\u0131fland\u0131rma i\u00e7in yap\u0131lan bir kestirimci modellemede, bu vekt\u00f6r tabanl\u0131 k\u0131yaslama, se\u00e7ilen \u00f6zelliklerin farkl\u0131 kategorilerin vekt\u00f6r uzay\u0131 i\u00e7inde birbirlerine uzak g\u00f6r\u00fcnd\u00fc\u011f\u00fc ayn\u0131 kategoriye ait olanlar\u0131n ise birbirine yak\u0131n g\u00f6r\u00fcnd\u00fc\u011f\u00fc durumlar\u0131 yaratmal\u0131d\u0131r. Bu ilke bir metin temsilini d\u00fczeltmek i\u00e7in de kullan\u0131labilir. Ayn\u0131 \u015fekilde s\u0131n\u0131fland\u0131r\u0131lmas\u0131 gereken durumlar\u0131 farkl\u0131 g\u00f6r\u00fcnmesini sa\u011flayan ya da farkl\u0131 \u015fekilde s\u0131n\u0131fland\u0131r\u0131lmas\u0131 gereken \u00f6rneklerin benzer g\u00f6r\u00fcnmesini sa\u011flayan \u00f6zellikler bu \u00f6zellikleri i\u00e7eren temsiller kullan\u0131larak e\u011fitilen modeller taraf\u0131nda yap\u0131lan s\u0131n\u0131fland\u0131rmalarda kuvvetle muhtemel bir kafa kar\u0131\u015f\u0131kl\u0131\u011f\u0131na neden olacakt\u0131r. Problem \u00e7o\u011funlukla ya mu\u011flak \u00f6zellikler (\u00f6r. Farkl\u0131 ba\u011flamlarda farkl\u0131 anlamlara gelebilen fakat g\u00f6sterimin bu ba\u011flam\u0131 mu\u011flakl\u0131\u011f\u0131 giderebilmek ad\u0131na bask\u0131lamas\u0131na imk\u00e2n vermeyen \u00f6zellikler) ya da par\u00e7alanmad\u0131r (\u00f6r. ayn\u0131 soyut \u00f6zellik bir\u00e7ok belirli \u00f6zellik taraf\u0131ndan da temsil ediliyor, baz\u0131lar\u0131 kay\u0131p veya verinizde \u00e7ok seyrek bulunuyorsa). Ayr\u0131ca, en anlaml\u0131 \u00f6zellikler \u00f6zellik alan\u0131ndan ve \u00f6\u011fretim verisi olarak kullan\u0131labilecek belirgin veriler i\u00e7indeki anlaml\u0131 verilerle ili\u015fkili olan di\u011fer verilerden eksik olabilir ve bu da model anlaml\u0131 \u00f6zellikler ile daha az anlaml\u0131 \u00f6zellikler aras\u0131ndaki sahte korelasyonlar\u0131n bulunmayabilece\u011fi veya farkl\u0131 olabilece\u011fi yeni verilere uyguland\u0131\u011f\u0131nda \u00fcretici kar\u015f\u0131t\u0131 olmakla sonu\u00e7lanabilecek \u015fekilde genelde \"dikkati \u00fczerinden \u00e7ekecektir\"<\/p>\n\n

Vaka Analizi<\/h3>\n

SA i\u00e7in metin temsili\/g\u00f6sterimine giren d\u00fc\u015f\u00fcn\u00fc\u015f bi\u00e7imini \u00f6rneklerle a\u00e7\u0131klamak i\u00e7in metinde tutum analizi3<\/sup><\/a> olarak bilinen, di\u011fer bir \u015fekilde de duygu analizi<\/i>4<\/sup><\/a> olarak da bilinen yayg\u0131n bir \u00f6rnekle ba\u015flayaca\u011f\u0131z (Pang veLee, 2008). Metin madencili\u011finin en yo\u011fun bir \u015fekilde pazarlanan uygulamalar\u0131ndan biridir ve ara\u015ft\u0131rmac\u0131lar\u0131n veriyi analiz etme durumunda metin verisine s\u0131kl\u0131kla ilk uygulamay\u0131 d\u00fc\u015f\u00fcnd\u00fckleri \u015feydir. Metin analitikleriyle ilgili baz\u0131 hususlar\u0131 tan\u0131tarak ba\u015flayaca\u011f\u0131z ve zorlanan ve sonunda at\u0131lma durumuna gelen \u00f6\u011frenciler taraf\u0131ndan makul olarak daha \u00e7ok olumsuz tutum ifadelerini g\u00f6rmeyi bekleyece\u011finiz KA\u00c7D\u2019lerdeki y\u0131pranma5<\/sup><\/a> \u00f6r\u00fcnt\u00fclerini a\u00e7\u0131klamak konusunda bu analitiklerin ne sundu\u011fu veya sunamad\u0131\u011f\u0131na dair bir inceleme ile bitirece\u011fiz. Resim bundan \u00e7ok daha karma\u015f\u0131k oldu\u011funu g\u00f6rece\u011fiz (Wen, Yang ve Ros\u00e9, 2014a). Okuyucuya bu vaka incelemesinde yol g\u00f6sterirken ki\u015finin fazlaca basitle\u015ftirilmi\u015f \u00f6nyarg\u0131lardan ba\u015flay\u0131p, tekrarlamalar sayesinde daha fazla bilgilenmi\u015f olarak veri analizi d\u00f6ng\u00fcleri boyunca nas\u0131l ilerlenebilece\u011fini anlayabilmesini umut ediyoruz. SA alan\u0131ndaki en ilgin\u00e7 \u00e7al\u0131\u015fma veya analitiklerin zengin ve g\u00f6rece olarak yap\u0131land\u0131r\u0131lm\u0131\u015f veriye uyguland\u0131\u011f\u0131 herhangi bir alan\u0131, benzer bir hik\u00e2ye kurgusunu takip edecektir.<\/p>\n

Duyguya dair basitle\u015ftirilmi\u015f i\u015flemler metinleri ya olumlu ya da olumsuz duygu sergileme \u015feklinde tan\u0131mlamakta ve kelimeler ile bu duyu\u015fsal yarg\u0131lar aras\u0131ndaki ba\u011fa g\u00fcvenmektedirler. Dolays\u0131yla i\u015fin \u00e7o\u011fu kelimelerin olumluluk ya da olumsuzluk puanlar\u0131yla ili\u015fkilendirildi\u011fi duygu s\u00f6zl\u00fckleri olu\u015fturmaya varmaktad\u0131r. Duygu analizi alan\u0131 iyi geli\u015ftirilmi\u015f bir aland\u0131r, sekt\u00f6rde \u00f6nemli \u00f6l\u00e7\u00fcde temsiliyet kazanmakta ve pazarlama konular\u0131yla ili\u015fkili i\u015f kollar\u0131na hizmet vermektedir. Yine de teknolojinin s\u0131n\u0131rl\u0131l\u0131klar\u0131 a\u00e7\u0131kt\u0131r. Ayr\u0131ca dil bilimsel alan yaz\u0131n\u0131ndan \u00f6\u011frenilen tutumlarla ilgili pek \u00e7ok \u015feyin belirli olumlu veya olumsuz s\u00f6zc\u00fcklerle ifade edilmedi\u011fi y\u00f6n\u00fcndedir (Martin ve White, 2005). Bu hava durumuna dair verilen \u015fu \u00f6rnekle a\u00e7\u0131klanabilir. \u201cBug\u00fcn hava \u00e7ok g\u00fczel\u201d ifadesi istenen olumlu s\u00f6zc\u00fc\u011f\u00fc i\u00e7erir; ancak \u201c g\u00fcne\u015f par\u0131ld\u0131yor\u201d sadece tipik g\u00fcne\u015fli g\u00fcnlerin ya\u011fmurlu g\u00fcnlere tercih edildi\u011fi biliniyorsa a\u00e7\u0131k\u00e7a olumludur. \u201c Kapal\u0131 mek\u00e2nlarda kalmak i\u00e7in harika bir g\u00fcn\u201d havan\u0131n, olumlu bir s\u00f6zc\u00fc\u011f\u00fcn varl\u0131\u011f\u0131na ra\u011fmen pekiyi olmad\u0131\u011f\u0131n\u0131 g\u00f6stermektedir. \u201c Ya\u011fmur botlar\u0131m unutulmu\u015f hissediyor\u201d olumsuz bir s\u00f6zc\u00fc\u011f\u00fcn varl\u0131\u011f\u0131na ra\u011fmen hava ile ilgili olumlu bir yorum olarak al\u0131nabilir.<\/p>\n

Art\u0131k eve daha yak\u0131n olan ve yetersiz kalan durumlar\u0131 ara\u015ft\u0131raca\u011f\u0131z. Duygu analizi metne ilgi duyan farkl\u0131 alanlardan ara\u015ft\u0131rmac\u0131lar ve uygulay\u0131c\u0131lar taraf\u0131ndan en \u00e7ok bilinen ve geni\u015f \u00f6l\u00e7\u00fcde kullan\u0131lan dil teknolojilerinden biri oldu\u011fundan, KA\u00c7D'lerden gelen forum verisinin analizinin bu teknolojinin uygulamalar\u0131n\u0131 buldu\u011fumuz alanlardan biri olmas\u0131 s\u00fcrpriz de\u011fildir ve bu sebeple bu \u00e7al\u0131\u015fma uygun bir vaka \u00e7al\u0131\u015fmas\u0131d\u0131r. Uygulaman\u0131n gerek\u00e7esi tart\u0131\u015fma forumu verilerinin \u00f6\u011frencilerin KA\u00c7D\u2019leri nas\u0131l neden ve ne zaman terk ettiklerini, dersten tatmin olmad\u0131klar\u0131 i\u00e7in ayr\u0131ld\u0131klar\u0131 ve bu tatminsizli\u011fin duygu analizini mercek olarak kullanarak g\u00f6r\u00fcn\u00fcr hale getirilebilece\u011fi fikrine dayanarak daha iyi anlamak i\u00e7in faydal\u0131 olaca\u011f\u0131yd\u0131. Ancak \u00f6nceden yap\u0131lan b\u00f6yle bir ara\u015ft\u0131rmada Ramesh, Goldwasser, Huang, Daum\u00e9 ve Getoor (2013) genel olarak \u00f6\u011frenciler taraf\u0131ndan ifade edilen duygu (tamamen otomatik y\u00f6ntemlerle de\u011ferlendirilen) ile ona ili\u015fkin dersi bitirme olas\u0131l\u0131klar\u0131 aras\u0131nda hi\u00e7bir ili\u015fki bulamam\u0131\u015flard\u0131r. Adamopoulos (2013) ders tart\u0131\u015fma forumlar\u0131nda \u00f6\u011frencilerin tutumlar\u0131na dair neyi ifade ettiklerini anlamak i\u00e7in farkl\u0131 derslerin olanaklar\u0131na ili\u015fkin duygunun \u00f6l\u00e7\u00fclmesine y\u00f6nelik duyguya dayal\u0131 bir de\u011ferlendirme metodu geli\u015ftirdi. Tutumlarla ba\u011flant\u0131l\u0131 olan ders boyutlar\u0131ndaki temalar\u0131 belirlemek amac\u0131yla, otomatik bir \u015fekilde tan\u0131mlanan duygu ifadelerinin temellendirilmi\u015f kuram yakla\u015f\u0131m\u0131 ile e\u015fle\u015ftirildi\u011fi bir birle\u015fim kulland\u0131lar. Bu daha ayr\u0131nt\u0131l\u0131 bak\u0131\u015fla, genel tutumun de\u011fil dersin \u00f6\u011fretmenine, \u00f6devler ve di\u011fer materyallere kar\u015f\u0131 tutumun dersin b\u0131rak\u0131lmas\u0131yla en g\u00fc\u00e7l\u00fc ili\u015fkiye sahip oldu\u011funu belirleyebildiler. Daha yak\u0131n zamanl\u0131 bir \u00e7al\u0131\u015fmada (Wen vd., 2014a), otomatik analizi, duygu \u00f6l\u00e7\u00fcm\u00fcnde kesinli\u011fi artt\u0131rarak ve bir \u00f6\u011frenci taraf\u0131ndan ifade edilen bir duygu ile maruz kald\u0131klar\u0131 duyguyu ve ayn\u0131 zamanda \u00f6\u011frenci d\u00fczeyinde duygu ile ders d\u00fczeyinde duyguyu kar\u015f\u0131la\u015ft\u0131rarak, bir ad\u0131m \u00f6teye ta\u015f\u0131d\u0131k. Bu \u00e7al\u0131\u015fmada, duyguya ili\u015fkin de\u011fi\u015fkenler ve dersi b\u0131rakma aras\u0131ndaki ger\u00e7ek ba\u011flant\u0131 dersin do\u011fas\u0131na g\u00f6re de\u011fi\u015fmi\u015ftir.<\/p>\n

Daha fazla soru sorulmas\u0131 ile g\u00f6nderilerdeki tutumu s\u0131n\u0131fland\u0131rmak i\u00e7in \u00e7ok daha incelikli yollara ihtiyac\u0131m\u0131z oldu\u011fu belirginle\u015fmi\u015ftir. \u00d6rne\u011fin, tamamen sosyal olan de\u011fi\u015f toku\u015flarda, olumsuz tutum ifadeleri peki\u015fmi\u015f duygusal ba\u011flant\u0131ya yol a\u00e7an bir if\u015fa etme durumu olabilir. Problem \u00e7\u00f6zme dersinde problem konu\u015fmas\u0131 tam da materyalle ba\u011flant\u0131ya ge\u00e7mi\u015f olman\u0131n belirtisi olabilir. Olumsuz tutum s\u00f6zc\u00fckleri, ifadeleri ve g\u00f6r\u00fcnt\u00fcleri talihsiz veya stresli olaylar\u0131n tart\u0131\u015f\u0131ld\u0131\u011f\u0131 bir edebiyat dersinde ortaya \u00e7\u0131kabilir ve yine bu, ifade edilen duygunun \u00f6\u011frencinin bu materyali okuma deneyimi ve hatta tart\u0131\u015fma deneyimine dair duygusu ile hi\u00e7 de ili\u015fkili olamayabilir. Duygu analizinin olumlu ya da olumsuz s\u00f6zc\u00fckleri saymak kadar basit olmad\u0131\u011f\u0131 sonucuna vard\u0131k. Bireysel s\u00f6zc\u00fckler tutum ve ba\u011flama ili\u015fkin i\u00e7in yeterli bir kan\u0131t olu\u015fturmamaktad\u0131rlar. Baz\u0131 retorik stratejiler olumlu ve olumsuz yorumlar\u0131 ayn\u0131 de\u011ferlendirme i\u00e7inde birle\u015ftirebilir ve bazen duygu dolayl\u0131 olarak ifade edilebilir. Verinizin g\u00f6steriminde g\u00f6zlem yoluyla yap\u0131lan nitel analizle bu \u015fekilde ince ayr\u0131nt\u0131lar mutlaka dikkate al\u0131nmal\u0131d\u0131r.<\/p>\n\n

DENET\u0130MS\u0130Z Y\u00d6NTEMLER<\/h2>\n

Fakt\u00f6r analizine ait \u00e7ok \u00e7e\u015fitli (Garson, 2013; Loehlin, 2004) ve \u00f6rt\u00fck de\u011fi\u015fken analizi teknikleri (Skrondal ve Rabe-Hesketh, 2004; Collins ve Lanza, 2010) bu alanda olduk\u00e7a pop\u00fclerdir. Bunlar denetimsiz (\u00f6nceden atanm\u0131\u015f etiketler gerektirmeyen vb.), denetlenen (\u00f6nceden atanm\u0131\u015f etiketleri olan \u00f6rnekleri gerektiren vb.), ya da az denetlenmi\u015f (\u00f6\u011frenme algoritmas\u0131 i\u00e7in biraz d\u0131\u015f rehberli\u011fe ihtiya\u00e7 duyulan fakat her \u00f6rnek i\u00e7in \u00f6nceden atanm\u0131\u015f etiketler gerektirmeyen) olabilirler. Bu b\u00f6l\u00fcmde denetimsiz y\u00f6ntemlere odaklanaca\u011f\u0131z. E\u011fitsel alandaki bu tekniklerin en pop\u00fcler olanlar\u0131ndan \u00f6rt\u00fck semantik analiz (\u00d6SA: Foltz, 1996) veya gizli Drichlet ayr\u0131m\u0131 veya GDT (Blei vd., 2003) gibi \u00f6rt\u00fck de\u011fi\u015fken fakt\u00f6r analiti\u011fi yakla\u015f\u0131mlar\u0131 yukar\u0131da k\u0131saca a\u00e7\u0131klanm\u0131\u015ft\u0131. Dolay\u0131s\u0131yla burada biraz daha ayr\u0131nt\u0131ya girecek ve g\u00fc\u00e7l\u00fc yanlar ile s\u0131n\u0131rl\u0131l\u0131klara de\u011finece\u011fiz. \u00d6A'ya ili\u015fkin yak\u0131n tarihli bir \u00e7al\u0131\u015fmada ke\u015ffedici veri analizi i\u00e7in denetimsiz y\u00f6ntemler (Joksimovi\u0107 vd., 2015; Sekiya, Marsuda ve Yamaguchi, 2015; Chen, Chen ve Xing, 2015), bazen g\u00f6rselle\u015ftirme teknikleri ile e\u015flenerek (Hsiao ve Awasthi, 2015) veya el analizine dayand\u0131r\u0131larak veya onunla de\u011fi\u015fmeli olarak (Molenaar ve Chiu, 2015; Ezen-Can, Boyer, Kellog ve Booth, 2015) kullan\u0131lm\u0131\u015ft\u0131r. Bu modelleme teknolojileri ara\u015ft\u0131rmac\u0131lar onlar\u0131 metinsel anlam analizine \u00e7ok yak\u0131n bulduklar\u0131 i\u00e7in yayg\u0131n olarak kullan\u0131lm\u0131\u015ft\u0131r. Ge\u00e7ek \u015fudur ki onlar bunu ger\u00e7ekle\u015ftirmek konusunda yayg\u0131n kan\u0131ya k\u0131yasla \u00e7ok daha az uygundurlar. Bu ara\u00e7lar ger\u00e7ekten SA ara\u00e7lar\u0131n\u0131n cephaneli\u011finde yer al\u0131rlar. Ancak bu b\u00f6l\u00fcm yukar\u0131da da belirtildi\u011fi gibi okuyucuda uygun bir \u015f\u00fcphecili\u011fi te\u015fvik etmek i\u00e7in biraz daha derinlere dalma merak\u0131n\u0131 uyand\u0131rmay\u0131 ama\u00e7lamaktad\u0131r.<\/p>\n

Etiketlenmemi\u015f verinin \u00e7ok \u00e7e\u015fitli bir\u00e7ok \u00f6zelli\u011fini modellemede konu modelleme yakla\u015f\u0131mlar\u0131 \u00e7ok pop\u00fclerdir. Belge toplaman\u0131n tematik yap\u0131s\u0131n\u0131 ortaya \u00e7\u0131karmada etkili olan \u00fcretici bir model olarak; iyi bilinen ve yayg\u0131n olarak kullan\u0131lan yakla\u015f\u0131mlardan birisi GDT\u2019d\u0131r (Blei vd., 2003). Sakl\u0131 Markov modelleme (SMM) ve di\u011fer dizilim modelleme yakla\u015f\u0131mlar\u0131 \u00f6\u011frenen deneyimlerindeki ilerlemeleri yakalamada pop\u00fcler hale gelmeye ba\u015flam\u0131\u015ft\u0131r (Molenaar ve Chiu, 2015). Bazen bu yakla\u015f\u0131mlar zamana g\u00f6re dil ifadelerinin yordanabilir yollarla ve tematik i\u00e7erik g\u00f6sterimleri cinsinden nas\u0131l de\u011fi\u015fti\u011finin belirlemek i\u00e7in birle\u015ftirilmektedir(Jo ve Ros\u00e9, 2015). Bunlar gibi istatistiksel yakla\u015f\u0131mlar\u0131n d\u00fczenlilikleri ortaya koymas\u0131 istenir. Metodolojilerde en \u00e7ok, veri azaltma ve basitle\u015ftirmeye de\u011fer katan ara\u00e7lar olarak de\u011ferlidirler. Veri i\u00e7indeki al\u0131\u015f\u0131lmad\u0131k olu\u015fumlar\u0131 \u00f6nemsiz g\u00f6rd\u00fckleri i\u00e7in varsay\u0131mlar\u0131 zorlayan s\u0131rad\u0131\u015f\u0131 olaylar\u0131 arayan metodolojilerde daha az de\u011ferlidirler. Ki\u015fi varsay\u0131mlar\u0131 ihlal eden durumlar\u0131 belirlemek i\u00e7in, b\u00f6yle \u00f6rnekleri belirlemenin bir yolu olarak bir anomali bulma yakla\u015f\u0131m\u0131n\u0131 se\u00e7se bile, uygulamada bulunan \u00f6rneklerin, teorik olarak \u00f6nemli olan varsay\u0131mlar\u0131 zorlamas\u0131 bak\u0131\u015f a\u00e7\u0131s\u0131na g\u00f6re tam da ilgin\u00e7 olmama ve s\u0131ra d\u0131\u015f\u0131 olmama ihtimali vard\u0131r.<\/p>\n

GDT kelimeleri ayn\u0131 belge i\u00e7erisinde s\u0131kl\u0131kla birlikte bulunan \u00f6rt\u00fck bir kelime s\u0131n\u0131f\u0131 i\u00e7inde birlikte e\u015fleme bi\u00e7iminde \u00e7al\u0131\u015fmaktad\u0131r. \u00d6\u011frenilen yakla\u015f\u0131m \u00f6rt\u00fck yap\u0131daki her bir veri nokta b\u00fct\u00fcn\u00fcn\u00fcn (bu bir belgedir), tek bir \u00f6rt\u00fck s\u0131n\u0131fa olas\u0131l\u0131\u011fa dayal\u0131 olarak atand\u0131\u011f\u0131 (Collins ve Lanza, 2010), geleneksel \u00f6rt\u00fck s\u0131n\u0131f modellerinden daha karma\u015f\u0131kt\u0131r. GDT modeline, veri noktalar\u0131n\u0131n \u00f6rt\u00fck s\u0131n\u0131flar\u0131n bir bile\u015fimi olarak g\u00f6r\u00fclerek belgelerin i\u00e7indeki s\u00f6zc\u00fcklerin faraz\u00ee olarak \u00f6rt\u00fck s\u0131n\u0131flara atand\u0131\u011f\u0131 ek bir yap\u0131 katman\u0131 daha konmu\u015ftur. Bu konu analizi i\u00e7in \u00f6nemli bir yap\u0131d\u0131r. Belge temsillerinin \u00f6rt\u00fck s\u00f6zc\u00fck s\u0131n\u0131flar\u0131n\u0131n herhangi bir bile\u015fimi olmas\u0131na izin vererek temalar\u0131n bireysel belgelerin i\u00e7inde harmanlanmas\u0131 esnekli\u011finin de korunmas\u0131yla \u00f6rt\u00fck s\u0131n\u0131flar\u0131n say\u0131s\u0131n\u0131n y\u00f6netilebilir bir boyutta kalmas\u0131 m\u00fcmk\u00fcn k\u0131l\u0131nabilir. Her bir \u00f6rt\u00fck s\u00f6zc\u00fck s\u0131n\u0131f\u0131 s\u00f6zc\u00fcklerin bir da\u011f\u0131l\u0131m\u0131yla temsil edilir. Da\u011f\u0131l\u0131mda en y\u00fcksek say\u0131da yer alan s\u00f6zc\u00fckler, ili\u015fkili \u00f6rt\u00fck s\u0131n\u0131f veya konunun en ay\u0131rt edici \u00f6zelli\u011fi olarak ele al\u0131n\u0131rlar.<\/p>\n

GDT'nin denetimsiz dil i\u015fleme tekniklerinden bir olmas\u0131ndan dolay\u0131 belirlenen temalar\u0131n konu temalar\u0131n\u0131n d\u00fczenlemesine dair insan \u00f6nsezisiyle tam olarak e\u015fle\u015fece\u011fini beklemek \u00e7ok da mant\u0131kl\u0131 olmayacakt\u0131r ancak s\u00f6zc\u00fcklerin birlikte bulunma ili\u015fkilerinin modellendi\u011fi bir teknik olarak ili\u015fkilendirilebilece\u011fi d\u00fc\u015f\u00fcn\u00fclen baz\u0131 \u015feylerin belirlenmesini beklemek m\u00fcmk\u00fcnd\u00fcr. GDT \u00f6z\u00fcnde bir veri azaltma tekni\u011fidir. G\u00fc\u00e7l\u00fc oldu\u011fu y\u00f6nler derlemde \u00e7ok yayg\u0131n olan, s\u0131kl\u0131kla ortak temalara kar\u015f\u0131l\u0131k gelen s\u00f6zc\u00fck ili\u015fkilendirmelerinin belirlemesinden kaynaklanmaktad\u0131r. Ancak ortak\/yayg\u0131n temalar\u0131n ilgilenilen temalarla bire bir uyum olmas\u0131 gerekmez. Ne yaz\u0131k ki, bu sonu\u00e7ta ortaya \u00e7\u0131kan g\u00f6sterimde ilgilenilen temalar\u0131n ortak\/yayg\u0131n olmayanlar\u0131 i\u00e7in farkl\u0131 bir g\u00f6sterim olu\u015fturulmayacakt\u0131r. Benzer olarak, yayg\u0131n kullan\u0131lan fikirlerin al\u0131\u015f\u0131lmad\u0131k ifadeleri genel olarak GDT sahas\u0131 i\u00e7inde bir sezgisel g\u00f6sterimle e\u015fle\u015fmeyi ba\u015faramayacakt\u0131r. Metin verisinin g\u00f6sterimi ayr\u0131 bir \u00f6neme sahiptir. S\u0131kl\u0131kla, GDT modelleri bireysel s\u00f6zc\u00fck niteliklerinden olu\u015fturulan nitelik alanlar\u0131 \u00fczerinden hesaplan\u0131r. B\u00f6ylelikle, bireysel s\u00f6zc\u00fcklerle yakalanmayan hi\u00e7bir \u015fey model taraf\u0131ndan eri\u015filebilir olmayacakt\u0131r.<\/p>\n\n

DENET\u0130ML\u0130 Y\u00d6NTEMLER<\/h2>\n

Spektrumun di\u011fer ucunda ise denetimli y\u00f6ntemler bulunmaktad\u0131r. Biraz fazlaca basitle\u015ftirilmi\u015f bir bak\u0131\u015f a\u00e7\u0131s\u0131yla ele al\u0131n\u0131rsa, denetimli makine \u00f6\u011frenimi y\u00f6ntemleri, genellikle s\u0131n\u0131f de\u011feri olarak adland\u0131r\u0131lan bir sonu\u00e7 \u00f6zelli\u011fi ile nitelik olarak adland\u0131r\u0131lan tahmin unsurlar\u0131ndan olu\u015fan bir koleksiyonu ili\u015fkilendirecek vekt\u00f6r k\u00fcmeleri \u00fczerinde \u00e7al\u0131\u015fan algoritmalard\u0131r. Son zamanlarda, denetimli makine \u00f6\u011frenmesinin \u00f6\u011frenme s\u00fcre\u00e7lerinin de\u011ferlendirilmesi problemine uygulanmas\u0131 tart\u0131\u015fma konudur. Bu problem otomatik- i\u015fbirlikli \u00f6\u011frenme s\u00fcre\u00e7 analizi olarak adland\u0131r\u0131lmaktad\u0131r. \u0130\u015fbirlikli s\u00fcre\u00e7lerin otomatik analizi, i\u015fbirlikli \u00f6\u011frenme s\u0131ras\u0131nda ger\u00e7ek zamanl\u0131 de\u011ferlendirme, i\u015fbirlikli-\u00f6\u011frenme oturumlar\u0131n\u0131n ortas\u0131nda destekleyici m\u00fcdahaleleri dinamik olarak tetikleme ve i\u015fbirlikli-\u00f6\u011frenme s\u00fcre\u00e7lerinin etkili bir \u015fekilde analiz edilmesini kolayla\u015ft\u0131rmak i\u00e7in de\u011fere sahiptir. Bu dinamik yakla\u015f\u0131m\u0131n di\u011fer e\u015fde\u011fer statik destek yakla\u015f\u0131m\u0131ndan daha etkili oldu\u011fu g\u00f6sterilmi\u015ftir Kumar, Ros\u00e9, Wang, Joshi ve Robinson, 2007). Otomatikle\u015ftirilmi\u015f i\u015fbirlikli \u00f6\u011frenme s\u00fcre\u00e7 analizindeki erken d\u00f6nem yap\u0131lm\u0131\u015f \u00e7al\u0131\u015fmalar metin temelli etkile\u015fimleri ve t\u0131klama dizisi verisine odaklanmaktayd\u0131 (Soller ve Lesgold, 2007; Erkens ve Janssen, 2008; Ros\u00e9 vd., 2008; McLaren vd., 2007; Mu vd., 2012). \u0130\u015fbirlikli s\u00fcre\u00e7lerin konu\u015fmayla analizine y\u00f6nelik eski \u00e7al\u0131\u015fmalar da ortaya \u00e7\u0131kmaktad\u0131r (Gweon vd., 2013; Gweon, Agarwal, Udani, Raj ve Ros\u00e9, 2011). Dilbilim ve psikolojideki teorik \u00e7er\u00e7evelerden hareketle olu\u015fturulmu\u015f g\u00f6sterimlerin umut vadetmesi ise tutarl\u0131 bir bulgudur (Ros\u00e9 ve Tovares, bas\u0131m a\u015famas\u0131nda; Wen, Yang ve Ros\u00e9, 2014b; Gweon vd., 2013; Ros\u00e9 ve VanLehn, 2005). Alanda deneyim edinmek i\u00e7in LightSIDE ara\u00e7 tezgah\u0131n\u0131n iyi bir ba\u015flang\u0131\u00e7 noktas\u0131 oldu\u011fundan daha \u00f6nce de bahsetmi\u015ftik.<\/p>\n\n

B\u0130R ADIM \u00d6TEYE GE\u00c7ERKEN<\/h2>\n

SA alan\u0131 ile ilgili daha fazla bilgi edinmeye istekli okuyucular ilk olarak, temel alan yaz\u0131na biraz dalmakla fayda sa\u011flayacaklard\u0131r. Bu dilbilim alan\u0131, (Levinson, 1983; O'Grady vd., 2009), s\u00f6ylem analizi (Martin ve Rose, 2003; Martin ve White, 2005; Biber ve Conrad, 2011) ve dil teknolojileri (Manning ve Schuetze, 1999; Jurafsky ve Martin, 2009; Jackson ve Moulinier, 2007) alanlar\u0131na dayand\u0131r\u0131lm\u0131\u015ft\u0131r.<\/p>\n\n

KAYNAK\u00c7A<\/h2>\nAdamopoulos, P. (2013). What makes a great MOOC? An interdisciplinary analysis of student retention in online courses. Proceedings of the 34th International Conference on Information Systems: Reshaping Society through Information Systems Design <\/i>(ICIS 2013), 15\u201318 December 2013, Milan, Italy. http:\/\/aisel.aisnet.org\/ icis2013\/proceedings\/BreakthroughIdeas\/13\/ <\/span>\n\nAllen, L., Snow, E., & McNamera, D. (2015). Are you reading my mind? Modeling students\u2019 reading comprehension skills with natural language processing techniques. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 246\u2013254). New York: ACM. <\/span>\n\nBiber, D., & Conrad, S. (2011). Register, Genre, and Style<\/i>. Cambridge, UK: Cambridge University Press. <\/span>\n\nBlei, D., Ng, A., & Jordan, M. (2003). Latent Dirichlet allocation. Journal of Machine Learning Research, 3<\/i>, 993\u20131022. <\/span>\n\nBuckingham Shum, S. (2013). Proceedings of the 1st International Workshop on Discourse-Centric Learning Analytics (DCLA13), 8 April 2013, Leuven, Belgium. <\/span>\n\nBuckingham Shum, S., de Laat, M., de Liddo, A., Ferguson, R., & Whitelock, D. (2014). Proceedings of the 2nd International Workshop on Discourse-Centric Learning Analytics <\/i>(DCLA14), 24 March 2014, Indianapolis, IN, USA. <\/span>\n\nChen, B., Chen, X., & Xing, W. (2015). \u201cTwitter archaeology\u201d of Learning Analytics and Knowledge conferences. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 340\u2013349). New York: ACM. <\/span>\n\nCollins, L., & Lanza, S. T. (2010). Latent class and latent transition analysis with applications in the social, behavioral, and health sciences. Wiley. <\/span>\n\nDascalu, M., Dessus, P., & McNamera, D. (2015). Discourse cohesion: A signature of collaboration. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 350\u2013354). New York: ACM. <\/span>\n\nErkens, G., & Janssen, J. (2008). Automatic coding of dialogue acts in collaboration protocols. International Journal of Computer-Supported Collaborative Learning, 3<\/i>, 447\u2013470. <\/span>\n\nEzen-Can, A., Boyer, K., Kellog, S., & Booth, S. (2015). Unsupervised modeling for understanding MOOC discussion forums: A learning analytics approach. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 146\u2013150). New York: ACM. <\/span>\n\nFoltz, P. (1996). Latent semantic analysis for text-based research. Behavior Research Methods, Instruments, & Computers, 28<\/i>(2), 197\u2013202. <\/span>\n\nGarson, G.D. (2013). Factor Analysis. Asheboro, NC: Statistical Associates Publishing. http:\/\/www.statisticalassociates.com\/factoranalysis.htm <\/span>\n\nGianfortoni, P., Adamson, D., & Ros\u00e9, C. P. (2011). Modeling stylistic variation in social media with stretchy patterns. Proceedings of the 1st Workshop on Algorithms and Resources for Modeling of Dialects and Language Varieties <\/i>(DIALECTS\u201911), 31 July 2011, Edinburgh, Scotland (pp. 49\u201359). Association for Computational Linguistics. <\/span>\n\nGriffiths, T. L., & Steyvers, M. (2004). Finding scientific topics. Proceedings of the National Academy of Sciences, 101<\/i>, 5228\u20135235. <\/span>\n\nGweon, G., Jain, M., McDonough, J., Raj, B., & Ros\u00e9, C. P. (2013). Measuring prevalence of other-oriented transactive contributions using an automated measure of speech style accommodation. International Journal of Computer Supported Collaborative Learning, 8<\/i>(2), 245\u2013265. <\/span>\n\nGweon, G., Agarwal, P., Udani, M., Raj, B., & Ros\u00e9, C. P. (2011). The automatic assessment of knowledge integration processes in project teams. Proceedings of the 9th International Conference on Computer-Supported Collaborative Learning, Volume 1: Long Papers <\/i>(CSCL 2011), 4\u20138 July 2011, Hong Kong, China (pp. 462\u2013469). International Society of the Learning Sciences.<\/span>\n\nHmelo-Silver, C., Chinn, C., Chan, C., & O\u2019Donnell, A. (2013). The International Handbook of Collaborative Learning<\/i>. Routledge. <\/span>\n\nHsiao, I., & Awasthi, P. (2015). Topic facet modeling: Semantic and visual analytics for online discussion forums. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 231\u2013235). New York: ACM. <\/span>\n\nJackson, P., & Moulinier, I. (2007). Natural language processing for online applications: Text retrieval, extraction, and categorization. Amsterdam: John Benjamins Publishing Company. <\/span>\n\nJo, Y., Loghmanpour, N., & Ros\u00e9, C. P. (2015). Time series analysis of nursing notes for mortality prediction via state transition topic models. Proceedings of the 24th ACM International Conference on Information and Knowledge Management <\/i>(CIKM\u201915), 19\u201323 October 2015, Melbourne, VIC, Australia (pp. 1171\u20131180). New York: ACM. <\/span>\n\nJoksimovi\u0107, S., Kovanovi\u0107, V., Jovanovi\u0107, J., Zouaq, A., Ga\u0161evi\u0107, D., & Hatala, M. (2015). What do cMOOC participants talk about in social media? A topic analysis of discourse in a cMOOC. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 156\u2013165). New York: ACM. <\/span>\n\nJurafsky, D., & Martin, J. (2009). Speech and language processing: An introduction to natural language processing, computational linguistics, and speech recognition. Pearson. <\/span>\n\nKnight, S., & Littleton, K. (2015). Developing a multiple-document-processing performance assessment for epistemic literacy. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 241\u2013245). New York: ACM. <\/span>\n\nKumar, R., Ros\u00e9, C. P., Wang, Y. C., Joshi, M., & Robinson, A. (2007). Tutorial dialogue as adaptive collaborative learning support. Proceedings of the 13th International Conference on Artificial Intelligence in Education: Building Technology Rich Learning Contexts that Work <\/i>(AIED 2007), 9\u201313 July 2007, Los Angeles, CA, USA (pp. 383\u2013390). IOS Press. <\/span>\n\nLevinson, S. (1983). Conversational structure. Pragmatics <\/i>(pp. 284\u2013286). Cambridge, UK: Cambridge University Press. <\/span>\n\nLoehlin, J.C. (2004). Latent variable models: An introduction to factor, path, and structural equation analysis. Routledge. <\/span>\n\nManning, C., & Schuetze, H. (1999). Foundations of statistical natural language processing. MIT Press. <\/span>\n\nMartin, J., & Rose, D. (2003). Working with discourse: Meaning beyond the clause. Continuum. <\/span>\n\nMartin, J., & White, P. (2005). The language of evaluation: Appraisal in English. Palgrave. <\/span>\n\nMayfield, E., & Ros\u00e9, C. P. (2013). LightSIDE: Open source machine learning for text accessible to non-experts. In M. D. Shermis & J. Burstein (Eds.), Handbook of Automated Essay Grading: Current Applications and New Directions <\/i>(pp. 124\u2013135). Routledge. <\/span>\n\nMcLaren, B., Scheuer, O., De Laat, M., Hever, R., de Groot, R., & Ros\u00e9, C. P. (2007). Using machine learning techniques to analyze and support mediation of student E-discussions. Proceedings of the 13th International Conference on Artificial Intelligence in Education: Building Technology Rich Learning Contexts That Work <\/i>(AIED 2007), 9\u201313 July 2007, Los Angeles, CA, USA (pp. 331\u2013338). IOS Press. <\/span>\n\nMcNamara, D. S., & Graesser, A. C. (2012). Coh-Metrix: An automated tool for theoretical and applied natural language processing. In P. M. McCarthy & C. Boonthum (Eds.), Applied Natural Language Processing: Identification, Investigation, and Resolution <\/i>(pp. 188\u2013205). Hershey, PA: IGI Global. <\/span>\n\nMilligan, S. (2015). Crowd-sourced learning in MOOCs: Learning analytics meets measurement theory. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 151\u2013155). New York: ACM. <\/span>\n\nMolenaar, I., & Chiu, M. (2015). Effects of sequences of socially regulated learning on group performance. Proceedings of the 5th Internation Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 236\u2013240). New York: ACM.<\/span>\n\nMorrow, R. A., & Brown, D. D. (1994). Deconstructing the conventional discourse of methodology: Quantitative versus qualitative methods. In R. A. Morrow & D. D. Brown (Eds.), Critical theory and methodology: Contemporary social theory<\/i>, Vol. 3 (pp. 199\u2013225). Thousand Oaks, CA: Sage. <\/span>\n\nMu, J., Stegmann, K., Mayfield, E., Ros\u00e9, C. P., & Fischer, F. (2012). The ACODEA framework: Developing segmentation and classification schemes for fully automatic analysis of online discussions. International Journal of Computer-Supported Collaborative Learning, 138<\/i>, 285\u2013305. <\/span>\n\nNehm, R., Ha, M., & Mayfeld, E. (2012). Transforming biology assessment with machine learning: Automated scoring of written evolutionary explanations. Journal of Science Education and Technology, 21<\/i>, 183\u2013196. <\/span>\n\nNguyen, D., Dogru\u00f6z, A. S., Ros\u00e9, C. P., & de Jong, F. (in press). Computational sociolinguistics: A survey. Computational Linguistics<\/i>. <\/span>\n\nO\u2019Donnell, A., & King, A. (1999). Cognitive perspectives on peer learning<\/i>. Routledge. <\/span>\n\nO\u2019Grady, W., Archibald, J., Aronoff, M., & Rees-Miller, J. (2009). Contemporary linguistics: An introduction<\/i>. Boston \/ New York: Bedford \/ St. Martin. <\/span>\n\nPage, E. B. (1966). The imminence of grading essays by computer. Phi Delta Kappan, 48<\/i>, 238\u2013243. <\/span>\n\nPang, B., & Lee, L. (2008). Opinion mining and sentiment analysis. Foundations and Trends in Information Retrieval, 2<\/i>(1\u20132), 1\u2013135. <\/span>\n\nRamesh, A., Goldwasser, D., Huang, B., Daum\u00e9 III, H., & Getoor, L. (2013). Modeling learner engagement in MOOCs using probabilistic soft logic. NIPS Workshop on Data Driven Education: Advances in Neural Information Processing Systems <\/i>(NIPS-DDE 2013), 9 December 2013, Lake Tahoe, NV, USA. https: \/\/www.umiacs. umd.edu\/~hal\/docs\/daume13engagementmooc.pdf <\/span>\n\nRibeiro, B.T. (2006). Footing, positioning, voice: Are we talking about the same things? In A. De Fina, D. Schiffrin, & M. Bamberg (Eds.), Discourse and identity <\/i>(pp. 48\u201382). New York: Cambridge University Press. <\/span>\n\nRos\u00e9, C. P., & Tovares, A. (2015). What sociolinguistics and machine learning have to say to one another about interaction analysis. In L. Resnick, C. Asterhan, & S. Clarke (Eds.), Socializing intelligence through academic talk and dialogue<\/i>. Washington, DC: American Educational Research Association. <\/span>\n\nRos\u00e9, C. P., & VanLehn, K. (2005). An evaluation of a hybrid language understanding approach for robust selection of tutoring goals. International Journal of Artificial Intelligence in Education, 15<\/i>, 325\u2013355. <\/span>\n\nRos\u00e9, C. P., Wang, Y. C., Cui, Y., Arguello, J., Stegmann, K., Weinberger, A., & Fischer, F., (2008). Analyzing collaborative learning processes automatically: Exploiting the advances of computational linguistics in computer-supported collaborative learning. International Journal of Computer Supported Collaborative Learning, 3(3), 237\u2013271. <\/span>\n\nSekiya, T., Marsuda, Y., & Yamaguchi, K. (2015). Curriculum analysis of CS departments based on CS2013 by simplified, supervised LDA. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 330\u2013339). New York: ACM. <\/span>\n\nShermis, M. D., & Burstein, J. (2013). Handbook of Automated Essay Evaluation: Current Applications and New Directions. New York: Routledge. <\/span>\n\nShermis, M., & Hammer, B. (2012). Contrasting state-of-the-art automated scoring of essays: Analysis. Annual National Council on Measurement in Education Meeting, 14\u201316. <\/span>\n\nSimsek, D., Sandor, A., & Buckingham Shum, S. (2015). Correlations between automated rhetorical analysis and tutor\u2019s grades on student essays. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 355\u2013359). New York: ACM. <\/span>\n\nSkrondal, A., & Rabe-Hesketh, S. (2004). Generalized latent variable modeling: Multi-level, longitudinal, and structural equation models. Chapman & Hall\/CRC. <\/span>\n\nSnow, E., Allen, L., Jacovina, M., Perret, C., & McNamera, D. (2015). You\u2019ve got style: Writing flexibility across time. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 194\u2013202). New York: ACM.<\/span>\n\nSoller, A., & Lesgold, A. (2007). Modeling the process of collaborative learning. In H. U. Hoppe, H. Ogata, & A. Soller (Eds.), The role of technology in CSCL: Studies in technology enhanced collaborative learning <\/i>(pp 63\u201386). Springer. doi:10.1007\/978-0-387-71136-2_5 <\/span>\n\nWen, M., Yang, D., & Ros\u00e9, C. P. (2014a). Sentiment analysis in MOOC discussion forums: What does it tell us? In J. Stamper, Z. Pardos, M. Mavrikis, & B. M. McLaren (Eds.), Proceedings of the 7th International Conference on Educational Data Mining <\/i>(EDM2014), 4\u20137 July, London, UK. International Educational Data Mining Society. https:\/\/www.researchgate.net\/publication\/264080975_Sentiment_analysis_in_MOOC_discussion_forums_What_does_it_tell_us <\/span>\n\nWen, M., Yang, D., & Ros\u00e9, C. P. (2014b). Linguistic reflections of student engagement in massive open online courses. Proceedings of the 8th International AAAI Conference on Weblogs and Social Media (ICWSM\u201914), 1\u20134 June 2014, Ann Arbor, Michigan, USA. Palo Alto, CA: AAAI Press. http:\/\/www.cs.cmu.edu\/~mwen\/papers\/ icwsm2014-camera-ready.pdf <\/span>\n\nWitten, I. H., Frank, E., & Hall, M. (2011). Data mining: Practical machine learning tools and techniques<\/i>, 3rd ed. San Francisco, CA: Elsevier.<\/span>\n
\n\n
\n\n1<\/a> https:\/\/www.edx.org\/course\/data\u2013analytics\u2013learning\u2013utarlingtonx\u2013link5\u201310x<\/span>\n\n<\/div>\n
\n\n2<\/a> http:\/\/lightsidelabs.com\/research\/<\/span>\n\n<\/div>\n
\n\n3<\/a> orj. affect analysis<\/span>\n\n<\/div>\n
\n\n4<\/a> orj. sentiment analysis <\/span>\n\n<\/div>\n
\n\n5<\/a> orj. attrition<\/span>\n\n<\/div>\n","rendered":"

Carolyn Penstein Rose<\/span><\/span><\/p>\n

Dil Teknolojileri Enstit\u00fcs\u00fc ve \u0130nsan-Bilgisayar Etkile\u015fimi Enstit\u00fcs\u00fc, Carnegie Mellon \u00dcniversitesi, ABD<\/span><\/span><\/p>\n

DOI: 10.18608\/hla17.009<\/span><\/span><\/p>\n

\u00d6Z<\/h2>\n

Bu b\u00f6l\u00fcm s\u00f6ylem analitiklerini (SA) tan\u0131tmaktad\u0131r. S\u00f6ylem analitikleri, ara\u015ft\u0131rmay\u0131 destekleyecek analitik mercekler sunmak, bi\u00e7imlendirici ve de\u011fer bi\u00e7meye y\u00f6nelik de\u011ferlendirmeleri i\u015fler k\u0131lmak, \u00f6\u011frenme etkinliklerinin etkilili\u011fini artt\u0131rmak amac\u0131yla yap\u0131lan m\u00fcdahaleleri dinamik ve ba\u011flama duyarl\u0131 bir \u015fekilde harekete ge\u00e7irmek ve \u00f6\u011frenme etkinliklerinden sonraki raporlar ve geri bildirimler gibi yans\u0131tma ara\u00e7lar\u0131n\u0131n hem \u00f6\u011frenmeyi hem de \u00f6\u011fretimi destekleyecek \u015fekilde temin edilmesi gibi bir\u00e7ok alana etki etmektedir. Bu b\u00f6l\u00fcm\u00fcn amac\u0131, bu alanda ne yap\u0131labilece\u011fine dair belirli bir miktar \u00fcmidi ve ku\u015fkuculu\u011fu y\u00fcreklendirmek ayn\u0131 zamanda okuyucuya anlaml\u0131 bir i\u015f yapabilmek i\u00e7in ihtiya\u00e7 duyulan derinlikte uzmanl\u0131\u011f\u0131 sunmakt\u0131r. Amac\u0131 gerekli uzmanl\u0131\u011f\u0131 vermek de\u011fildir. Aksine, burada ama\u00e7 okuyucunun yeterli derinli\u011fi sa\u011flayacak bir ekip olu\u015fturmak i\u00e7in ne t\u00fcr i\u015fbirlikli \u00e7al\u0131\u015fma ortaklar\u0131 arayaca\u011f\u0131n\u0131 kavramada kendi yerini belirlemesidir. Alan\u0131n bir tan\u0131mlamas\u0131yla ba\u015flay\u0131p hem teorik hem de metodolojik olarak geni\u015f bir alan\u0131 ku\u015fat\u0131p hem temsil hem algoritmik boyutlar\u0131n\u0131 ara\u015ft\u0131racak ve daha derinlere dalmaya niyetli okuyuculara sonraki ad\u0131m \u00f6nerileriyle bitirece\u011fiz.<\/span><\/p>\n

Anahtar kelimeler<\/span>: S\u00f6ylem analizi, i\u015fbirlikli \u00f6\u011frenme, makine \u00f6\u011frenmesi, analiz ara\u00e7lar\u0131<\/span><\/p>\n

S\u00f6ylem analiti\u011fi (SA) \u00f6\u011frenme analitikleri i\u00e7inde bir aland\u0131r (\u00d6A; Buckingham Shum, 2013; Buckingham Shum, de Laat, de Liddo, Ferguson ve Whitelock, 2014). E\u011fitsel ortamlar i\u00e7indeki a\u00e7\u0131k u\u00e7lu sorular\u0131n i\u015flenmesini kapsar ve alandaki ara\u015ft\u0131rmalar b\u00fcy\u00fck oranda yaz\u0131 \u00e7al\u0131\u015fmalar\u0131n\u0131n de\u011ferlendirilmesi konusuna odaklanm\u0131\u015ft\u0131r ancak bundan daha fazlas\u0131n\u0131; tart\u0131\u015fma forumlar\u0131nda, sohbet odalar\u0131nda, mikro bloglarda, bloglarda ve hatta wikilerde yap\u0131lan tart\u0131\u015fmalar\u0131 da kapsar. \u00d6A\u2019y\u0131 genel olarak, \u00f6\u011frenenlerin \u00f6\u011frenmelerini dinleme yoluyla \u00f6\u011frenmeyi \u00f6\u011frenme olarak ele al\u0131r, dinleyi\u015fimiz ise \u00e7o\u011funlukla veri madencili\u011fi ve makine \u00f6\u011frenmesi teknolojileri ile desteklenir ancak alanda yay\u0131nlanm\u0131\u015f \u00e7al\u0131\u015fmalar \u00f6nc\u00fc olsa da t\u00fcm durumlar bir otomasyon s\u00f6z konusu de\u011fildir (Knight ve Littleton, 2015; Milligan, 2015). Ayr\u0131ca, biz bu alan\u0131 farkl\u0131 k\u0131lan \u015feyin, verinin \u00fcretildi\u011fi t\u00fcm ak\u0131\u015flar\u0131nda, dinlemeye ait olan do\u011fal dil verisine odaklan\u0131lmas\u0131 oldu\u011funu d\u00fc\u015f\u00fcnmekteyiz.<\/p>\n

Bu b\u00f6l\u00fcm \u00d6A sahas\u0131 i\u00e7inde konumlanm\u0131\u015f olan bu alana \u00e7ok temel bir ba\u015flang\u0131\u00e7 niteli\u011findedir. SA de\u011fi\u015fimli olarak iki tehlikeli kavram yan\u0131lg\u0131s\u0131ndan muzdariptir. Birincisi asl\u0131nda bir\u00e7ok ki\u015fi i\u00e7in, kullan\u0131ma haz\u0131r olan ve analiz i\u015fini bir d\u00fc\u011fmeye basarak onlar ad\u0131na yapacak bir \u00e7\u00f6z\u00fcme sahip olma arzusuyla k\u00f6r\u00fcklenen a\u015f\u0131r\u0131 ve u\u00e7 bir beklentidir. Bu kavram yan\u0131lg\u0131s\u0131n\u0131n tutsa\u011f\u0131 olanlar hayal k\u0131r\u0131kl\u0131\u011f\u0131na u\u011framaya mahk\u00fbmdur. En basit veya en g\u00fc\u00e7l\u00fc modelleme teknolojilerinin etkili kullan\u0131m\u0131 \u00e7ok\u00e7a haz\u0131rl\u0131k, emek ve uzmanl\u0131k gerektirmektedir. \u0130kinci kavram yan\u0131lg\u0131s\u0131 ise bazen birinci kavram yan\u0131lg\u0131s\u0131ndan do\u011fan hayal k\u0131r\u0131kl\u0131klar\u0131n\u0131 veya s\u00f6ylemin karma\u015f\u0131kl\u0131klar\u0131n\u0131 son derece derinlikli bir \u015fekilde bilmekten kaynaklanan, hi\u00e7bir bilgisayar\u0131n var olan ince ayr\u0131nt\u0131lar\u0131 tamamen yakalayamayaca\u011f\u0131 fikrini yok sayman\u0131n zorlu\u011fu sonucu olarak ortaya \u00e7\u0131km\u0131\u015f olan a\u015f\u0131r\u0131 bir ku\u015fkuculuktur. S\u00f6ylem inan\u0131lmaz bir \u015fekilde karma\u015f\u0131k olsa da teknoloji harikas\u0131 modelleme yakla\u015f\u0131mlar\u0131n\u0131n tan\u0131mlayabildi\u011fi anlaml\u0131 \u00f6r\u00fcnt\u00fcler oldu\u011fu da bir ger\u00e7ektir. Bu b\u00f6l\u00fcm boyunca teknolojinin bug\u00fcnk\u00fc durumunu a\u00e7\u0131klayan \u00d6\u011frenme Analiti\u011fi, Bilgi ve di\u011fer ba\u011flant\u0131l\u0131 konferanslardan al\u0131nan bir\u00e7ok yay\u0131mlanm\u0131\u015f \u00e7al\u0131\u015fmaya at\u0131fta bulunulmu\u015ftur. Bilgi i\u015flemsel sosyo dilbilime dair yap\u0131lan yak\u0131n zamanl\u0131 bir ara\u015ft\u0131rma, hik\u00e2yeyi dil teknolojileri dal\u0131n\u0131n perspektifinden anlatmaktad\u0131r (Nguyen, Dogru\u00f6z, Ros\u00e9 ve de Jong, bas\u0131m a\u015famas\u0131nda) ve konuya \u00f6zel ilgi duyan okuyucular\u0131n ilgisini \u00e7ekecektir.<\/p>\n

Bu b\u00f6l\u00fcm, biraz daha derine dalmay\u0131 isteyen okuyuculara faydal\u0131 ipu\u00e7lar\u0131 sunmay\u0131 umut etmektedir. SA konusu ile ilgili ge\u00e7mi\u015f iki at\u00f6lye \u00e7al\u0131\u015fmas\u0131 \u00d6A toplulu\u011fu i\u00e7indeki temel \u00e7al\u0131\u015fmalar\u0131 ara\u015ft\u0131rm\u0131\u015ft\u0131r (Buckingham Shum, 2013; Buckingham Shum vd., 2014). Daha daralt\u0131lm\u0131\u015f kapsamda Bilgisayar destekli i\u015fbirlikli \u00f6\u011frenme ile ilgili konu ve y\u00f6ntemleri i\u00e7eren kapsaml\u0131 bir genel de\u011ferlendirme daha \u00f6nce hakemli dergilerde bas\u0131lm\u0131\u015f \u00fc\u00e7 makalede bulunabilir (Ros\u00e9 vd., 2008; Mu, Stegman, Mayfield, Ros\u00e9 ve Fischer, 2012; Gweon, Jain, McDonough, Raj ve Ros\u00e9, 2013). Alana dair k\u0131sa bir ders ise, edX platformunda verilen 2014 Veri Analiti\u011fi ve KA\u00c7D’yi \u00d6\u011frenme1<\/sup><\/a> deki metin madencili\u011fi b\u00f6l\u00fcm\u00fcnde bulunabilir. Di\u011fer kaynaklar b\u00f6l\u00fcm\u00fcn sonunda sunulacakt\u0131r.<\/p>\n

Bu b\u00f6l\u00fcmde; \u00f6\u011frenme olaylar\u0131 s\u0131ras\u0131nda ifade edilen do\u011fal dille ilgileniyoruz. Kuramsal ve metodolojik olarak kapsay\u0131c\u0131 olmak istiyoruz. Teorik ve metodolojik olarak kapsay\u0131c\u0131 olmay\u0131 arzu ediyoruz. S\u00f6ylem analitikleri ile ilgili yap\u0131lm\u0131\u015f h\u00e2lihaz\u0131rdaki \u00e7al\u0131\u015fmalar\u0131n \u00e7o\u011fu, \u00f6\u011frenmeyi ve onun dil ile ba\u011flant\u0131s\u0131n\u0131 bili\u015fsel bir mercekten g\u00f6rmektedir, ba\u015fka bir deyi\u015fle, s\u00f6ylem i\u00e7erisinde var olan dil davran\u0131\u015f\u0131 kategorileri aray\u0131\u015f\u0131; ortak s\u00f6ylem s\u00fcre\u00e7leri ve \u00f6\u011frenme ile ili\u015fkili olan bili\u015fsel s\u00fcre\u00e7ler aras\u0131ndaki ba\u011flant\u0131dan dolay\u0131 \u00f6\u011frenme kazan\u0131mlar\u0131 ile ilgili baz\u0131 \u00f6ng\u00f6r\u00fclerde bulunmaktad\u0131r. Bu b\u00f6l\u00fcmde \u00f6\u011frenme ve onun dille olan ba\u011flant\u0131s\u0131n\u0131, \u00f6\u011frenmede rol\u00fc olan bili\u015fsel ve sosyal fakt\u00f6rler aras\u0131ndaki \u00f6nemli etkile\u015fimi g\u00fc\u00e7lendirmek ad\u0131na sosyal bir mercek arac\u0131l\u0131\u011f\u0131 ile g\u00f6rmeyi ama\u00e7l\u0131yoruz. (Hmelo-Silver, Chinn, Chan ve O\u2019Donnell, 2013; O\u2019Donnell ve King, 1999). \u00d6rne\u011fin, \u00f6\u011frenme etkile\u015fimlerinde \u00f6nemli bir destekleyici rol oynayan temel e\u011filimler, tutumlar ve ili\u015fkileri ortaya \u00e7\u0131karacak s\u00f6ylem s\u00fcre\u00e7lerini belirlemeyi ama\u00e7l\u0131yoruz. Hangi durumda ifade edilmi\u015f olursa olsun do\u011fal dil son derece ki\u015fisel ve k\u00fclt\u00fcreldir. \u0130\u00e7erisine ki\u015fisel deneyimlerimizin ve bizden \u00f6nceki ku\u015faklar\u0131n yap\u0131tlar\u0131 yerle\u015fmi\u015f durumdad\u0131r. Dil se\u00e7imlerimizdeki detaylar bilin\u00e7li olarak yans\u0131tt\u0131\u011f\u0131m\u0131z, ayn\u0131 zamanda bilin\u00e7li olarak saklad\u0131\u011f\u0131m\u0131z ve hatta fark\u0131nda bile olmad\u0131\u011f\u0131m\u0131z kimliklerimiz hakk\u0131nda ipu\u00e7lar\u0131 verir. Hedef kitlemize dair ve onlara y\u00f6nelik tutumumuz ve hedef kitlemize g\u00f6re kendimizi konumland\u0131r\u0131\u015f\u0131m\u0131z hakk\u0131ndaki varsay\u0131mlar\u0131m\u0131z\u0131 veya bazen sadece hedef kitlemizin bizim yapt\u0131\u011f\u0131m\u0131z\u0131 d\u00fc\u015f\u00fcnmelerini istedi\u011fimiz varsay\u0131mlar\u0131 yans\u0131t\u0131rlar. Biz bu se\u00e7imleri ili\u015fkiler ekonomisi i\u00e7erisinde, benimsedi\u011fimiz hedeflere ula\u015fmak i\u00e7in bir para birimi gibi kullan\u0131r\u0131z (Ribeiro, 2006).<\/p>\n

Bu anlay\u0131\u015fla hesaplamay\u0131, \u00f6\u011frenenleri dinlemeyi desteklemek i\u00e7in bir mercek olarak kullan\u0131rken, bizimle \u00f6\u011frenme s\u00fcre\u00e7leri aras\u0131nda duran teknolojilere her daim; bir t\u00fcr dijital bi\u00e7ime kay\u0131t yap\u0131l\u0131rken neyin kayboldu\u011fu ve neyin d\u00f6n\u00fc\u015ft\u00fc\u011f\u00fc, hatta analitik teknolojinin uygulanmas\u0131 s\u0131ras\u0131nda ger\u00e7ekle\u015fen sonraki indirgeme ve d\u00f6n\u00fc\u015f\u00fcm de d\u00e2hil olmak \u00fczere yorumlama konusundaki sorumlulu\u011fumuzun bir k\u0131sm\u0131n\u0131; b\u0131rak\u0131yor oldu\u011fumuzu kabul etmemiz gerekir (Morrow ve Brown, 1994). Bu \u015ferhi de d\u00fc\u015ferek, bu b\u00f6l\u00fcmde yo\u011fun bir \u015fekilde model yorumlama ve ge\u00e7erlik de\u011ferlendirmesine dair sorulara odaklanaca\u011f\u0131z.<\/p>\n

BU B\u00d6L\u00dcM\u00dcN KAPSAMI VE ODA\u011eI<\/h2>\n

Herhangi bir ki\u015fi analitikleri d\u00fc\u015f\u00fcnd\u00fc\u011f\u00fcnde, zihninde aniden algoritmalar belirir (Witten, Frank ve Hall, 2011). Ancak uygulamal\u0131 istatistikten ders \u00e7\u0131karmak ve onun yerine \u00f6nce g\u00f6sterimler hakk\u0131nda d\u00fc\u015f\u00fcnmek \u00f6nem arz eder. SA \u00e7al\u0131\u015fmas\u0131n\u0131n kalbinde verinin g\u00f6sterimine odaklanmak yatar. Makine \u00f6\u011frenmesi modelleri do\u011frudan metinlere uygulanamaz. Daha ziyade, metinden kestirim \u00f6zellikleri elde edilmelidir. Bu \u00f6ng\u00f6r\u00fcc\u00fc \u00f6zellikler, sorular olarak alg\u0131lanabilir: \u201cMetinde_var m\u0131?\u201d veya \u201cMetinde_ka\u00e7 kez bulunuyor?\u201d Her bir \u00f6zellik bu sorulardan biriyse, her durumda, \u00f6zellik de\u011feri, sorunun cevab\u0131d\u0131r. \u0130lgilenen okuyucular; kamuya a\u00e7\u0131k LightSIDE arac\u01312<\/sup><\/a> (Mayfield ve Ros\u00e9, 2013; Gianfortoni, Adamson ve Ros\u00e9, 2011), gibi geni\u015f bir kullan\u0131c\u0131 k\u0131lavuzu, \u00f6rnek veri k\u00fcmeleri, s\u00fcre\u00e7le ilgili y\u00f6nergeler ve yard\u0131m etmeye haz\u0131r ara\u015ft\u0131rmac\u0131lar\u0131n ileti\u015fim bilgilerini i\u00e7eren; \u00fccretsiz olarak eri\u015filebilen, kullan\u0131ma haz\u0131r bir tezg\u00e2h (workbench) ile deneyler yaparak metinden elde edilebilecek basit \u00f6zelliklerin derinli\u011fine ve s\u0131n\u0131flama modellerinin yordama kesinli\u011fi \u00fczerinde nas\u0131l bir etkisi oldu\u011funa dair iyi bir fikir edinebilirler.<\/p>\n

Metne uygulanan modelleme tekniklerinde ba\u015far\u0131n\u0131n anahtar\u0131 anlaml\u0131 ipu\u00e7lar\u0131 \u00fcretecek do\u011fru sorular\u0131 sormakt\u0131r. Bu soruya ili\u015fkin d\u00fc\u015f\u00fcnme dilin nas\u0131l yap\u0131land\u0131r\u0131ld\u0131\u011f\u0131n\u0131 dikkate almakla ba\u015flar. Y\u00fczeysel olarak \u00e7\u0131plak g\u00f6ze dil yekpare, yap\u0131land\u0131r\u0131lmam\u0131\u015f bir b\u00fct\u00fcn olarak g\u00f6r\u00fcnse de asl\u0131nda \u00e7oklu katmanlardan olu\u015fan, her biri dilbilimin ayr\u0131 bir alan\u0131n\u0131n i\u00e7inde tan\u0131mlanan bir yap\u0131dan olu\u015fur. Bir dilbilim ders kitab\u0131n\u0131n (O\u2019Grady, Archibald, Aronoff, & Rees \u2013 Miller, 2009) giri\u015f niteli\u011findeki bir incelemesi, bu \u00d6A alan\u0131na girmek isteyen ara\u015ft\u0131rmac\u0131lar i\u00e7in de\u011ferli bir kaynak olacakt\u0131r. En k\u00fc\u00e7\u00fck par\u00e7ac\u0131\u011f\u0131 ses yap\u0131s\u0131 d\u00fczeyindedir, fonoloji (ses bilim) olarak bilinmektedir. Burada dilin temel ses b\u00f6l\u00fcmleri ve bunlar\u0131n dilin hece yap\u0131s\u0131na nas\u0131l uyum sa\u011flad\u0131\u011f\u0131 tan\u0131mlanmaktad\u0131r. Seslerin temel bir alfabesi bir dizi ses birimlerini olu\u015fturur fakat leh\u00e7eler i\u00e7inde bunlar belirli \u015fekillerde telaffuz edilebilirler ve bu etnisite, sosyoekonomik d\u00fczey ve b\u00f6lge gibi sosyal anlaml\u0131l\u0131kla ilgili olan de\u011fi\u015fkenler k\u00fcmesi ile olan ba\u011flant\u0131s\u0131ndan dolay\u0131 sosyal anlamda bir \u00f6nem ta\u015f\u0131maktad\u0131r. Bu d\u00fczeyin hem en \u00fcst\u00fcnde, morfoloji (bi\u00e7im bilim) olarak bilinen kelimelerin daha i\u00e7 yap\u0131s\u0131n\u0131n tan\u0131mland\u0131\u011f\u0131 bir katman vard\u0131r. Buras\u0131 dil bilgisi derslerinde \u00f6\u011frendi\u011fimiz ve bu arada fillerin zamanlar\u0131n\u0131 veya isimlerin say\u0131lar\u0131n\u0131 de\u011fi\u015ftiren tak\u0131lar\u0131n ortaya \u00e7\u0131kt\u0131\u011f\u0131 yerdir. Yukar\u0131da, t\u00fcm c\u00fcmlelerin gramer yap\u0131s\u0131n\u0131n tan\u0131mland\u0131\u011f\u0131 s\u00f6zdizimi d\u00fczeyidir. Ayn\u0131 zamanda c\u00fcmle d\u00fczeyinde anlam\u0131n de\u011fi\u015fmez ifadeler yoluyla, kurallar taraf\u0131ndan ve s\u00f6z dizim kurallar\u0131 taraf\u0131ndan y\u00f6nlendirilen daha k\u00fc\u00e7\u00fck b\u00f6l\u00fcmler d\u00fczenlenerek olu\u015fturuldu\u011fu anlam bilim alan\u0131 vard\u0131r ve s\u00f6zc\u00fcksel anlam bilim d\u00fczeyinde alt d\u00fczey anlam bilim b\u00f6l\u00fcmleriyle ili\u015fkilidir. C\u00fcmle seviyesinin \u00fcst\u00fcnde, yap\u0131n\u0131n di\u011fer y\u00f6nleri aras\u0131nda retorik stratejileri buldu\u011fumuz s\u00f6ylem d\u00fczeyidir. Bu teknik terimler bir\u00e7ok okuyucuya yabanc\u0131 gelebilir ancak daha ileri okumalar i\u00e7in uygun kaynaklar\u0131 bulmak isteyen okuyuculara yararl\u0131 arama terimleri sunacaklard\u0131r.<\/p>\n

Do\u011fal dil verisinin otomatikle\u015ftirilmi\u015f analizin hedefi oldu\u011fu bir\u00e7ok alan\u0131n tarihini izlersek, ge\u00e7erli modelleme i\u00e7in kilit unsurun anlaml\u0131 g\u00f6sterimler tasarlamak olarak adland\u0131r\u0131ld\u0131\u011f\u0131, ayn\u0131 nakarat\u0131 duyar\u0131z. Bu \u00f6rne\u011fi bu b\u00f6l\u00fcme koymakta, okuyucular\u0131n ayn\u0131 dersi zorlu bir \u015fekilde \u00f6\u011frenmekten korunmalar\u0131 \u00fcmidi yatmaktad\u0131r. SA ilgili olan bu dersin iyi \u00f6\u011frenildi\u011fi en eski durumlardan biri otomatikle\u015ftirilmi\u015f kompozisyon yaz\u0131s\u0131 puanlama ile ilgiliydi (Page, 1966; Shermis ve Hammer, 2012). En eski yakla\u015f\u0131mlar regresyon gibi basit modelleri ve ortalama c\u00fcmle uzunlu\u011fu, uzun kelime say\u0131s\u0131n\u0131 ve kompozisyonun uzunlu\u011funu sayma gibi basit \u00f6zellikleri kulland\u0131lar. Bu yakla\u015f\u0131mlar, say\u0131sal puanlar\u0131n atanmas\u0131n\u0131n g\u00fcvenilirli\u011fi a\u00e7\u0131s\u0131ndan olduk\u00e7a ba\u015far\u0131l\u0131 olmu\u015ftur (Shermis ve Burstein, 2013); Ancak de\u011ferlendirme i\u00e7in kan\u0131t kullan\u0131m\u0131nda ge\u00e7erlili\u011fi olmad\u0131\u011f\u0131 i\u00e7in ele\u015ftirildiler. Sonraki \u00e7al\u0131\u015fmalarda, odak noktas\u0131 daha \u00e7ok \u00f6\u011fretenlerin yazmay\u0131 puanlad\u0131klar\u0131 kendi r\u00fcbriklerine (dereceli puanlama anahtar\u0131) neleri d\u00e2hil ettikleri gibi \u00f6zelliklerin belirlenmesine kayd\u0131. Bu inceleme genellikle metni harf harf b\u00f6len (unigram) dil g\u00f6sterimlerini temel ald\u0131klar\u0131 i\u00e7in unigram \u00f6zelliklerle ilgili problemlerin kurban\u0131 olsalar da yine de i\u00e7erik tabanl\u0131 de\u011ferlendirmeleri desteklemek amac\u0131yla fakt\u00f6r analizine benzeyen \u00f6rt\u00fck semantik analiz (latent semantic analysis) (\u00d6SA: Foltz, 1996) veya gizli Drichlet tahsisi (latent Drichlet allocation) (GDT; Blei, Ng ve Jordan, 2003; Griffiths ve Steyvers, 2004) gibi teknikleri de kapsayarak i\u00e7erik odakl\u0131 \u00f6zelliklerin d\u00e2hil edilmesine neden oldu. CohMeTrix (McNamara ve Graesser, 2012) gibi di\u011fer fakt\u00f6r analitik dil analizi yakla\u015f\u0131mlar\u0131, bili\u015fsel g\u00fc\u00e7l\u00fck gibi fakt\u00f6rler d\u00e2hil \u00e7e\u015fitli boyutlar\u0131n yan\u0131nda, \u00f6\u011frencilerin yaz\u0131lar\u0131n\u0131n de\u011ferlendirilmesinde kullan\u0131lm\u0131\u015ft\u0131r. Belli bir d\u00fczeyde s\u00f6z dizimsel yap\u0131sal analizleri kullanan son derece g\u00fcndelik \u00e7al\u0131\u015fma alanlar\u0131nda CohMetrix yararlar sa\u011flam\u0131\u015ft\u0131r(Ros\u00e9 ve VanLehn, 2005). Fen e\u011fitiminde a\u00e7\u0131k u\u00e7lu sorular\u0131n de\u011ferlendirilmesinde LightSIDE ile ba\u015far\u0131 elde edilmi\u015ftir (Nehm, Ha ve Mayfield, 2012; Mayfield ve Ros\u00e9, 2013).<\/p>\n

Bu noktada SA\u2019ya dair a\u015f\u0131r\u0131 ve d\u00fc\u015f\u00fck beklentiler aras\u0131ndaki gerilime geri d\u00f6nmek faydal\u0131 olacakt\u0131r. Uygun ve anlaml\u0131 \u00f6zelliklerin belirlenmesindeki zorluklar\u0131 d\u00fc\u015f\u00fcnd\u00fc\u011f\u00fcm\u00fczde modelleme ara\u00e7lar\u0131 arac\u0131l\u0131\u011f\u0131 ile olu\u015fturdu\u011fumuz merceklerin s\u0131n\u0131rl\u0131l\u0131klar\u0131 ile uzla\u015fmam\u0131z gereklidir. SA ara\u015ft\u0131rmac\u0131lar veya uygulay\u0131c\u0131lar\u0131n ellerinde, kendileri ile d\u0131\u015f d\u00fcnyada ger\u00e7ekle\u015fen \u00f6\u011frenme par\u00e7alar\u0131 aras\u0131nda duran bir mercek olarak hizmet verir ya da \u00f6\u011frenen ve \u00f6\u011fretenler, \u00f6\u011frenenler, ya da \u00f6\u011frenen ile \u00f6\u011frenme teknolojileri aras\u0131nda bir filtre olabilirler. Mercekler kendileri arac\u0131l\u0131\u011f\u0131 ile g\u00f6r\u00fclen d\u00fcnyan\u0131n b\u00fct\u00fcn ayr\u0131nt\u0131lar\u0131n\u0131 basit\u00e7e aktarmad\u0131klar\u0131 i\u00e7in kesinlikle faydal\u0131d\u0131rlar. Aksine g\u00f6r\u00fcnt\u00fclerin onlar olmadan etkili bir \u015fekilde g\u00f6r\u00fclemeyecek \u00f6zelliklerini vurgularlar. Bu da onlar\u0131n yapmas\u0131na ihtiya\u00e7 duydu\u011fumuz \u015feydir. Bunlar\u0131 yapmak i\u00e7in ihtiyac\u0131m\u0131z olan \u015fey budur. Ayn\u0131 zamanda, tasar\u0131m taraf\u0131ndan daha az ilgin\u00e7 olarak kabul edilen \u00f6zellikleri ise karart\u0131rlar. Mercekler her zaman e\u011fip b\u00fckerler. Fakat onlar\u0131 ge\u00e7erli bir \u015fekilde kullanmak i\u00e7in, uygun bir mercek se\u00e7ebilmek ad\u0131na her birinin neyi vurgulad\u0131\u011f\u0131n\u0131 ya da neyi karartt\u0131\u011f\u0131n\u0131 ve g\u00f6rd\u00fc\u011f\u00fcm\u00fcz \u015feyi ge\u00e7erli bir \u015fekilde yorumlayabilmemiz i\u00e7in resmin onsuz veya ba\u015fka bir mercekle nas\u0131l olabilece\u011fini her zaman sorgulayarak anlamam\u0131z \u015fartt\u0131r. Bu y\u00fczden biz, en ba\u015f\u0131ndan itibaren bu alandaki ara\u015ft\u0131rmay\u0131 kullananlar, bu mercekleri geli\u015ftirenler veya onlar\u0131 etkin olarak ara\u015ft\u0131rmada veya uygulamada kullananlar\u0131 uygulama s\u0131ras\u0131nda neyin ka\u00e7\u0131n\u0131lmaz olarak kayboldu\u011fu ya da d\u00f6n\u00fc\u015ft\u00fc\u011f\u00fc konusunda kontroll\u00fc olmalar\u0131 konusunda uyar\u0131yoruz. Bu b\u00f6l\u00fcm dikkatini SA’n\u0131n kapsam\u0131nda olan \u00f6zel alanlara y\u00f6neltecektir.<\/p>\n

MET\u0130N G\u00d6STER\u0130M\u0130<\/h2>\n

Verinin analitik mercekler arac\u0131l\u0131\u011f\u0131 ile nas\u0131l g\u00f6r\u00fcnece\u011fini fazlas\u0131yla etkileyecek \u00f6nemli kararlar temsil a\u015famas\u0131nda al\u0131n\u0131r. Bu a\u015famada, metin yekpare g\u00f6r\u00fcnen bir b\u00fct\u00fcnden onun i\u00e7inden \u00e7\u0131kar\u0131labilecek bir dizi belirleyici niteli\u011fe d\u00f6n\u00fc\u015ft\u00fcr\u00fcl\u00fcr. Her \u00f6zellik \u00e7\u0131kar\u0131c\u0131, metne bir soru ve metnin verdi\u011fi cevap ise g\u00f6sterimde ona tekab\u00fcl eden niteli\u011fin de\u011feridir. Bir insan hakk\u0131nda tek bildi\u011finiz \u015feyin yirmi soruluk bir oyunda sorulan sorulara verilen bir dizi cevap oldu\u011funu hayal edin ve \u015fimdi g\u00f6reviniz bu insan\u0131 bir tak\u0131m sosyal g\u00f6r\u00fc\u015f kategorileri i\u00e7inde s\u0131n\u0131fland\u0131rmak olsun. E\u011fer sorular dikkatlice yap\u0131land\u0131r\u0131lm\u0131\u015fsa, hatas\u0131z bir kestirimde bulunabilirsiniz ama yine de bu ki\u015fiye ait bir\u00e7ok bilgi ve \u00f6ng\u00f6r\u00fcn\u00fcn s\u00fcre\u00e7 i\u00e7erisinde kaybolaca\u011f\u0131n\u0131 da kabul etmeniz gerekir. Bu \u00f6nemli a\u015famada bilgi bir kez kayboldu\u011funda hik\u00e2ye, dil teknolojileri alan\u0131n\u0131n bak\u0131\u015f a\u00e7\u0131s\u0131yla ne kadar ileri d\u00fczey olursa olsun bir algoritman\u0131n uygulanmas\u0131 ile yeniden elde edilemez. Bu nedenle, bu b\u00f6l\u00fcm boyunca temsil ve g\u00f6sterimlerle ilgili kararlar\u0131n dikkatlice al\u0131nmas\u0131n\u0131n, yorum \u00fczerinde dikkatlice ve etrafl\u0131ca d\u00fc\u015f\u00fcnmenin ve yapt\u0131\u011f\u0131m\u0131z \u00e7\u0131kar\u0131mlar\u0131n ge\u00e7erlili\u011fini dikkatlice sorgulaman\u0131n \u00f6nemini vurguluyoruz. Bu alanda yeni olan okuyucular bu ikazlar\u0131 hayali bulabilirlerse de deneyimle bu daha netle\u015feceklerdir.<\/p>\n

Genel Bak\u0131\u015f<\/h3>\n

Metni harf harf b\u00f6lme (unigram) \u00f6zellikleri metin madencili\u011fi problemlerinde kullan\u0131lan en tipik \u00f6zellik \u00e7\u0131kar\u0131c\u0131lar\u0131d\u0131r. Bir unigram \u00f6zellik aral\u0131\u011f\u0131nda e\u011fitim verilerindeki bir dizi metnin i\u00e7inde g\u00f6r\u00fclen her kelime i\u00e7in bu kelimenin metnin i\u00e7indeki varl\u0131\u011f\u0131n\u0131 soracak uygun bir \u00f6zellik olacakt\u0131r. Unigram \u00f6zellik aral\u0131klar\u0131 genellikle makul bir \u015fekilde y\u00fcksek performans elde ederken, modeller e\u011fitim verilerininkine \u00e7ok benzer ko\u015fullar alt\u0131nda toplanan verilerin \u00f6tesine genelleme yapmakta ba\u015far\u0131s\u0131zd\u0131rlar. Genellemedeki yetersizli\u011fin nedeni bu unigram modellerinin temel olarak her s\u0131n\u0131f de\u011fer etiketini y\u00fczeysel bir \u015fekilde ezberlemesidir. Bu modeller insanlar\u0131n belirli bir dizi durumda nelerden bahsettiklerini e\u011fitim verilerinde bulunan bu etiketle ili\u015fkilendirirler. E\u011fer bunda bir tutarl\u0131l\u0131k varsa, bu o zaman modeller taraf\u0131ndan \u00f6\u011frenilebilir ancak bu tutarl\u0131l\u0131k nadiren daha \u00f6teye ge\u00e7ip genelle\u015fir. \u00d6zellikler ortaya \u00e7\u0131kar\u0131ld\u0131\u011f\u0131nda olu\u015fan genelleme, amaca uygun bir yap\u0131 katman\u0131ndan gelir.<\/p>\n

Metnin \u00f6zellik tabanl\u0131 g\u00f6steriminin amac\u0131, m\u00fcmk\u00fcn olan en y\u00fcksek do\u011frulukla kestirimci modellemeyi ger\u00e7ekle\u015ftirmek hedefiyle, s\u0131kl\u0131kla s\u0131n\u0131fland\u0131rma veya say\u0131sal de\u011ferlendirme amac\u0131yla kestirimci modellemeyi etkinle\u015ftirmektir (Ros\u00e9 vd., 2008; Mc-Laren vd., 2007; Allen, Snow, McNamera, 2015). Bu b\u00f6l\u00fcm\u00fcn oda\u011f\u0131 ise uyumland\u0131rma s\u00fcreci olacakt\u0131r. Ancak SA’n\u0131n geni\u015f alan\u0131 i\u00e7inde bulunan baz\u0131 \u00e7al\u0131\u015fmalarda, temsil ve g\u00f6sterimler odak noktas\u0131d\u0131r, anlam belirlenen \u00f6ng\u00f6r\u00fcsel \u00f6zellikten \u00e7\u0131kar\u0131l\u0131r ve kestirimsel\/ \u00f6ng\u00f6r\u00fcsel modellemenin, e\u011fer varsa, belirlenen \u00f6zelliklerin bir do\u011frulamas\u0131 olarak hizmet etti\u011finin dikkate al\u0131nmas\u0131 \u00f6nemlidir (Simsek, Sandor ve Buckingham Shum, 2015; Dascalu, Dessus, McNamera, 2015; Snow, Allen, Jacovina, Perret, McNamera, 2015).<\/p>\n

S\u0131n\u0131fland\u0131rma i\u00e7in yap\u0131lan bir kestirimci modellemede, bu vekt\u00f6r tabanl\u0131 k\u0131yaslama, se\u00e7ilen \u00f6zelliklerin farkl\u0131 kategorilerin vekt\u00f6r uzay\u0131 i\u00e7inde birbirlerine uzak g\u00f6r\u00fcnd\u00fc\u011f\u00fc ayn\u0131 kategoriye ait olanlar\u0131n ise birbirine yak\u0131n g\u00f6r\u00fcnd\u00fc\u011f\u00fc durumlar\u0131 yaratmal\u0131d\u0131r. Bu ilke bir metin temsilini d\u00fczeltmek i\u00e7in de kullan\u0131labilir. Ayn\u0131 \u015fekilde s\u0131n\u0131fland\u0131r\u0131lmas\u0131 gereken durumlar\u0131 farkl\u0131 g\u00f6r\u00fcnmesini sa\u011flayan ya da farkl\u0131 \u015fekilde s\u0131n\u0131fland\u0131r\u0131lmas\u0131 gereken \u00f6rneklerin benzer g\u00f6r\u00fcnmesini sa\u011flayan \u00f6zellikler bu \u00f6zellikleri i\u00e7eren temsiller kullan\u0131larak e\u011fitilen modeller taraf\u0131nda yap\u0131lan s\u0131n\u0131fland\u0131rmalarda kuvvetle muhtemel bir kafa kar\u0131\u015f\u0131kl\u0131\u011f\u0131na neden olacakt\u0131r. Problem \u00e7o\u011funlukla ya mu\u011flak \u00f6zellikler (\u00f6r. Farkl\u0131 ba\u011flamlarda farkl\u0131 anlamlara gelebilen fakat g\u00f6sterimin bu ba\u011flam\u0131 mu\u011flakl\u0131\u011f\u0131 giderebilmek ad\u0131na bask\u0131lamas\u0131na imk\u00e2n vermeyen \u00f6zellikler) ya da par\u00e7alanmad\u0131r (\u00f6r. ayn\u0131 soyut \u00f6zellik bir\u00e7ok belirli \u00f6zellik taraf\u0131ndan da temsil ediliyor, baz\u0131lar\u0131 kay\u0131p veya verinizde \u00e7ok seyrek bulunuyorsa). Ayr\u0131ca, en anlaml\u0131 \u00f6zellikler \u00f6zellik alan\u0131ndan ve \u00f6\u011fretim verisi olarak kullan\u0131labilecek belirgin veriler i\u00e7indeki anlaml\u0131 verilerle ili\u015fkili olan di\u011fer verilerden eksik olabilir ve bu da model anlaml\u0131 \u00f6zellikler ile daha az anlaml\u0131 \u00f6zellikler aras\u0131ndaki sahte korelasyonlar\u0131n bulunmayabilece\u011fi veya farkl\u0131 olabilece\u011fi yeni verilere uyguland\u0131\u011f\u0131nda \u00fcretici kar\u015f\u0131t\u0131 olmakla sonu\u00e7lanabilecek \u015fekilde genelde “dikkati \u00fczerinden \u00e7ekecektir”<\/p>\n

Vaka Analizi<\/h3>\n

SA i\u00e7in metin temsili\/g\u00f6sterimine giren d\u00fc\u015f\u00fcn\u00fc\u015f bi\u00e7imini \u00f6rneklerle a\u00e7\u0131klamak i\u00e7in metinde tutum analizi3<\/sup><\/a> olarak bilinen, di\u011fer bir \u015fekilde de duygu analizi<\/i>4<\/sup><\/a> olarak da bilinen yayg\u0131n bir \u00f6rnekle ba\u015flayaca\u011f\u0131z (Pang veLee, 2008). Metin madencili\u011finin en yo\u011fun bir \u015fekilde pazarlanan uygulamalar\u0131ndan biridir ve ara\u015ft\u0131rmac\u0131lar\u0131n veriyi analiz etme durumunda metin verisine s\u0131kl\u0131kla ilk uygulamay\u0131 d\u00fc\u015f\u00fcnd\u00fckleri \u015feydir. Metin analitikleriyle ilgili baz\u0131 hususlar\u0131 tan\u0131tarak ba\u015flayaca\u011f\u0131z ve zorlanan ve sonunda at\u0131lma durumuna gelen \u00f6\u011frenciler taraf\u0131ndan makul olarak daha \u00e7ok olumsuz tutum ifadelerini g\u00f6rmeyi bekleyece\u011finiz KA\u00c7D\u2019lerdeki y\u0131pranma5<\/sup><\/a> \u00f6r\u00fcnt\u00fclerini a\u00e7\u0131klamak konusunda bu analitiklerin ne sundu\u011fu veya sunamad\u0131\u011f\u0131na dair bir inceleme ile bitirece\u011fiz. Resim bundan \u00e7ok daha karma\u015f\u0131k oldu\u011funu g\u00f6rece\u011fiz (Wen, Yang ve Ros\u00e9, 2014a). Okuyucuya bu vaka incelemesinde yol g\u00f6sterirken ki\u015finin fazlaca basitle\u015ftirilmi\u015f \u00f6nyarg\u0131lardan ba\u015flay\u0131p, tekrarlamalar sayesinde daha fazla bilgilenmi\u015f olarak veri analizi d\u00f6ng\u00fcleri boyunca nas\u0131l ilerlenebilece\u011fini anlayabilmesini umut ediyoruz. SA alan\u0131ndaki en ilgin\u00e7 \u00e7al\u0131\u015fma veya analitiklerin zengin ve g\u00f6rece olarak yap\u0131land\u0131r\u0131lm\u0131\u015f veriye uyguland\u0131\u011f\u0131 herhangi bir alan\u0131, benzer bir hik\u00e2ye kurgusunu takip edecektir.<\/p>\n

Duyguya dair basitle\u015ftirilmi\u015f i\u015flemler metinleri ya olumlu ya da olumsuz duygu sergileme \u015feklinde tan\u0131mlamakta ve kelimeler ile bu duyu\u015fsal yarg\u0131lar aras\u0131ndaki ba\u011fa g\u00fcvenmektedirler. Dolays\u0131yla i\u015fin \u00e7o\u011fu kelimelerin olumluluk ya da olumsuzluk puanlar\u0131yla ili\u015fkilendirildi\u011fi duygu s\u00f6zl\u00fckleri olu\u015fturmaya varmaktad\u0131r. Duygu analizi alan\u0131 iyi geli\u015ftirilmi\u015f bir aland\u0131r, sekt\u00f6rde \u00f6nemli \u00f6l\u00e7\u00fcde temsiliyet kazanmakta ve pazarlama konular\u0131yla ili\u015fkili i\u015f kollar\u0131na hizmet vermektedir. Yine de teknolojinin s\u0131n\u0131rl\u0131l\u0131klar\u0131 a\u00e7\u0131kt\u0131r. Ayr\u0131ca dil bilimsel alan yaz\u0131n\u0131ndan \u00f6\u011frenilen tutumlarla ilgili pek \u00e7ok \u015feyin belirli olumlu veya olumsuz s\u00f6zc\u00fcklerle ifade edilmedi\u011fi y\u00f6n\u00fcndedir (Martin ve White, 2005). Bu hava durumuna dair verilen \u015fu \u00f6rnekle a\u00e7\u0131klanabilir. \u201cBug\u00fcn hava \u00e7ok g\u00fczel\u201d ifadesi istenen olumlu s\u00f6zc\u00fc\u011f\u00fc i\u00e7erir; ancak \u201c g\u00fcne\u015f par\u0131ld\u0131yor\u201d sadece tipik g\u00fcne\u015fli g\u00fcnlerin ya\u011fmurlu g\u00fcnlere tercih edildi\u011fi biliniyorsa a\u00e7\u0131k\u00e7a olumludur. \u201c Kapal\u0131 mek\u00e2nlarda kalmak i\u00e7in harika bir g\u00fcn\u201d havan\u0131n, olumlu bir s\u00f6zc\u00fc\u011f\u00fcn varl\u0131\u011f\u0131na ra\u011fmen pekiyi olmad\u0131\u011f\u0131n\u0131 g\u00f6stermektedir. \u201c Ya\u011fmur botlar\u0131m unutulmu\u015f hissediyor\u201d olumsuz bir s\u00f6zc\u00fc\u011f\u00fcn varl\u0131\u011f\u0131na ra\u011fmen hava ile ilgili olumlu bir yorum olarak al\u0131nabilir.<\/p>\n

Art\u0131k eve daha yak\u0131n olan ve yetersiz kalan durumlar\u0131 ara\u015ft\u0131raca\u011f\u0131z. Duygu analizi metne ilgi duyan farkl\u0131 alanlardan ara\u015ft\u0131rmac\u0131lar ve uygulay\u0131c\u0131lar taraf\u0131ndan en \u00e7ok bilinen ve geni\u015f \u00f6l\u00e7\u00fcde kullan\u0131lan dil teknolojilerinden biri oldu\u011fundan, KA\u00c7D’lerden gelen forum verisinin analizinin bu teknolojinin uygulamalar\u0131n\u0131 buldu\u011fumuz alanlardan biri olmas\u0131 s\u00fcrpriz de\u011fildir ve bu sebeple bu \u00e7al\u0131\u015fma uygun bir vaka \u00e7al\u0131\u015fmas\u0131d\u0131r. Uygulaman\u0131n gerek\u00e7esi tart\u0131\u015fma forumu verilerinin \u00f6\u011frencilerin KA\u00c7D\u2019leri nas\u0131l neden ve ne zaman terk ettiklerini, dersten tatmin olmad\u0131klar\u0131 i\u00e7in ayr\u0131ld\u0131klar\u0131 ve bu tatminsizli\u011fin duygu analizini mercek olarak kullanarak g\u00f6r\u00fcn\u00fcr hale getirilebilece\u011fi fikrine dayanarak daha iyi anlamak i\u00e7in faydal\u0131 olaca\u011f\u0131yd\u0131. Ancak \u00f6nceden yap\u0131lan b\u00f6yle bir ara\u015ft\u0131rmada Ramesh, Goldwasser, Huang, Daum\u00e9 ve Getoor (2013) genel olarak \u00f6\u011frenciler taraf\u0131ndan ifade edilen duygu (tamamen otomatik y\u00f6ntemlerle de\u011ferlendirilen) ile ona ili\u015fkin dersi bitirme olas\u0131l\u0131klar\u0131 aras\u0131nda hi\u00e7bir ili\u015fki bulamam\u0131\u015flard\u0131r. Adamopoulos (2013) ders tart\u0131\u015fma forumlar\u0131nda \u00f6\u011frencilerin tutumlar\u0131na dair neyi ifade ettiklerini anlamak i\u00e7in farkl\u0131 derslerin olanaklar\u0131na ili\u015fkin duygunun \u00f6l\u00e7\u00fclmesine y\u00f6nelik duyguya dayal\u0131 bir de\u011ferlendirme metodu geli\u015ftirdi. Tutumlarla ba\u011flant\u0131l\u0131 olan ders boyutlar\u0131ndaki temalar\u0131 belirlemek amac\u0131yla, otomatik bir \u015fekilde tan\u0131mlanan duygu ifadelerinin temellendirilmi\u015f kuram yakla\u015f\u0131m\u0131 ile e\u015fle\u015ftirildi\u011fi bir birle\u015fim kulland\u0131lar. Bu daha ayr\u0131nt\u0131l\u0131 bak\u0131\u015fla, genel tutumun de\u011fil dersin \u00f6\u011fretmenine, \u00f6devler ve di\u011fer materyallere kar\u015f\u0131 tutumun dersin b\u0131rak\u0131lmas\u0131yla en g\u00fc\u00e7l\u00fc ili\u015fkiye sahip oldu\u011funu belirleyebildiler. Daha yak\u0131n zamanl\u0131 bir \u00e7al\u0131\u015fmada (Wen vd., 2014a), otomatik analizi, duygu \u00f6l\u00e7\u00fcm\u00fcnde kesinli\u011fi artt\u0131rarak ve bir \u00f6\u011frenci taraf\u0131ndan ifade edilen bir duygu ile maruz kald\u0131klar\u0131 duyguyu ve ayn\u0131 zamanda \u00f6\u011frenci d\u00fczeyinde duygu ile ders d\u00fczeyinde duyguyu kar\u015f\u0131la\u015ft\u0131rarak, bir ad\u0131m \u00f6teye ta\u015f\u0131d\u0131k. Bu \u00e7al\u0131\u015fmada, duyguya ili\u015fkin de\u011fi\u015fkenler ve dersi b\u0131rakma aras\u0131ndaki ger\u00e7ek ba\u011flant\u0131 dersin do\u011fas\u0131na g\u00f6re de\u011fi\u015fmi\u015ftir.<\/p>\n

Daha fazla soru sorulmas\u0131 ile g\u00f6nderilerdeki tutumu s\u0131n\u0131fland\u0131rmak i\u00e7in \u00e7ok daha incelikli yollara ihtiyac\u0131m\u0131z oldu\u011fu belirginle\u015fmi\u015ftir. \u00d6rne\u011fin, tamamen sosyal olan de\u011fi\u015f toku\u015flarda, olumsuz tutum ifadeleri peki\u015fmi\u015f duygusal ba\u011flant\u0131ya yol a\u00e7an bir if\u015fa etme durumu olabilir. Problem \u00e7\u00f6zme dersinde problem konu\u015fmas\u0131 tam da materyalle ba\u011flant\u0131ya ge\u00e7mi\u015f olman\u0131n belirtisi olabilir. Olumsuz tutum s\u00f6zc\u00fckleri, ifadeleri ve g\u00f6r\u00fcnt\u00fcleri talihsiz veya stresli olaylar\u0131n tart\u0131\u015f\u0131ld\u0131\u011f\u0131 bir edebiyat dersinde ortaya \u00e7\u0131kabilir ve yine bu, ifade edilen duygunun \u00f6\u011frencinin bu materyali okuma deneyimi ve hatta tart\u0131\u015fma deneyimine dair duygusu ile hi\u00e7 de ili\u015fkili olamayabilir. Duygu analizinin olumlu ya da olumsuz s\u00f6zc\u00fckleri saymak kadar basit olmad\u0131\u011f\u0131 sonucuna vard\u0131k. Bireysel s\u00f6zc\u00fckler tutum ve ba\u011flama ili\u015fkin i\u00e7in yeterli bir kan\u0131t olu\u015fturmamaktad\u0131rlar. Baz\u0131 retorik stratejiler olumlu ve olumsuz yorumlar\u0131 ayn\u0131 de\u011ferlendirme i\u00e7inde birle\u015ftirebilir ve bazen duygu dolayl\u0131 olarak ifade edilebilir. Verinizin g\u00f6steriminde g\u00f6zlem yoluyla yap\u0131lan nitel analizle bu \u015fekilde ince ayr\u0131nt\u0131lar mutlaka dikkate al\u0131nmal\u0131d\u0131r.<\/p>\n

DENET\u0130MS\u0130Z Y\u00d6NTEMLER<\/h2>\n

Fakt\u00f6r analizine ait \u00e7ok \u00e7e\u015fitli (Garson, 2013; Loehlin, 2004) ve \u00f6rt\u00fck de\u011fi\u015fken analizi teknikleri (Skrondal ve Rabe-Hesketh, 2004; Collins ve Lanza, 2010) bu alanda olduk\u00e7a pop\u00fclerdir. Bunlar denetimsiz (\u00f6nceden atanm\u0131\u015f etiketler gerektirmeyen vb.), denetlenen (\u00f6nceden atanm\u0131\u015f etiketleri olan \u00f6rnekleri gerektiren vb.), ya da az denetlenmi\u015f (\u00f6\u011frenme algoritmas\u0131 i\u00e7in biraz d\u0131\u015f rehberli\u011fe ihtiya\u00e7 duyulan fakat her \u00f6rnek i\u00e7in \u00f6nceden atanm\u0131\u015f etiketler gerektirmeyen) olabilirler. Bu b\u00f6l\u00fcmde denetimsiz y\u00f6ntemlere odaklanaca\u011f\u0131z. E\u011fitsel alandaki bu tekniklerin en pop\u00fcler olanlar\u0131ndan \u00f6rt\u00fck semantik analiz (\u00d6SA: Foltz, 1996) veya gizli Drichlet ayr\u0131m\u0131 veya GDT (Blei vd., 2003) gibi \u00f6rt\u00fck de\u011fi\u015fken fakt\u00f6r analiti\u011fi yakla\u015f\u0131mlar\u0131 yukar\u0131da k\u0131saca a\u00e7\u0131klanm\u0131\u015ft\u0131. Dolay\u0131s\u0131yla burada biraz daha ayr\u0131nt\u0131ya girecek ve g\u00fc\u00e7l\u00fc yanlar ile s\u0131n\u0131rl\u0131l\u0131klara de\u011finece\u011fiz. \u00d6A’ya ili\u015fkin yak\u0131n tarihli bir \u00e7al\u0131\u015fmada ke\u015ffedici veri analizi i\u00e7in denetimsiz y\u00f6ntemler (Joksimovi\u0107 vd., 2015; Sekiya, Marsuda ve Yamaguchi, 2015; Chen, Chen ve Xing, 2015), bazen g\u00f6rselle\u015ftirme teknikleri ile e\u015flenerek (Hsiao ve Awasthi, 2015) veya el analizine dayand\u0131r\u0131larak veya onunla de\u011fi\u015fmeli olarak (Molenaar ve Chiu, 2015; Ezen-Can, Boyer, Kellog ve Booth, 2015) kullan\u0131lm\u0131\u015ft\u0131r. Bu modelleme teknolojileri ara\u015ft\u0131rmac\u0131lar onlar\u0131 metinsel anlam analizine \u00e7ok yak\u0131n bulduklar\u0131 i\u00e7in yayg\u0131n olarak kullan\u0131lm\u0131\u015ft\u0131r. Ge\u00e7ek \u015fudur ki onlar bunu ger\u00e7ekle\u015ftirmek konusunda yayg\u0131n kan\u0131ya k\u0131yasla \u00e7ok daha az uygundurlar. Bu ara\u00e7lar ger\u00e7ekten SA ara\u00e7lar\u0131n\u0131n cephaneli\u011finde yer al\u0131rlar. Ancak bu b\u00f6l\u00fcm yukar\u0131da da belirtildi\u011fi gibi okuyucuda uygun bir \u015f\u00fcphecili\u011fi te\u015fvik etmek i\u00e7in biraz daha derinlere dalma merak\u0131n\u0131 uyand\u0131rmay\u0131 ama\u00e7lamaktad\u0131r.<\/p>\n

Etiketlenmemi\u015f verinin \u00e7ok \u00e7e\u015fitli bir\u00e7ok \u00f6zelli\u011fini modellemede konu modelleme yakla\u015f\u0131mlar\u0131 \u00e7ok pop\u00fclerdir. Belge toplaman\u0131n tematik yap\u0131s\u0131n\u0131 ortaya \u00e7\u0131karmada etkili olan \u00fcretici bir model olarak; iyi bilinen ve yayg\u0131n olarak kullan\u0131lan yakla\u015f\u0131mlardan birisi GDT\u2019d\u0131r (Blei vd., 2003). Sakl\u0131 Markov modelleme (SMM) ve di\u011fer dizilim modelleme yakla\u015f\u0131mlar\u0131 \u00f6\u011frenen deneyimlerindeki ilerlemeleri yakalamada pop\u00fcler hale gelmeye ba\u015flam\u0131\u015ft\u0131r (Molenaar ve Chiu, 2015). Bazen bu yakla\u015f\u0131mlar zamana g\u00f6re dil ifadelerinin yordanabilir yollarla ve tematik i\u00e7erik g\u00f6sterimleri cinsinden nas\u0131l de\u011fi\u015fti\u011finin belirlemek i\u00e7in birle\u015ftirilmektedir(Jo ve Ros\u00e9, 2015). Bunlar gibi istatistiksel yakla\u015f\u0131mlar\u0131n d\u00fczenlilikleri ortaya koymas\u0131 istenir. Metodolojilerde en \u00e7ok, veri azaltma ve basitle\u015ftirmeye de\u011fer katan ara\u00e7lar olarak de\u011ferlidirler. Veri i\u00e7indeki al\u0131\u015f\u0131lmad\u0131k olu\u015fumlar\u0131 \u00f6nemsiz g\u00f6rd\u00fckleri i\u00e7in varsay\u0131mlar\u0131 zorlayan s\u0131rad\u0131\u015f\u0131 olaylar\u0131 arayan metodolojilerde daha az de\u011ferlidirler. Ki\u015fi varsay\u0131mlar\u0131 ihlal eden durumlar\u0131 belirlemek i\u00e7in, b\u00f6yle \u00f6rnekleri belirlemenin bir yolu olarak bir anomali bulma yakla\u015f\u0131m\u0131n\u0131 se\u00e7se bile, uygulamada bulunan \u00f6rneklerin, teorik olarak \u00f6nemli olan varsay\u0131mlar\u0131 zorlamas\u0131 bak\u0131\u015f a\u00e7\u0131s\u0131na g\u00f6re tam da ilgin\u00e7 olmama ve s\u0131ra d\u0131\u015f\u0131 olmama ihtimali vard\u0131r.<\/p>\n

GDT kelimeleri ayn\u0131 belge i\u00e7erisinde s\u0131kl\u0131kla birlikte bulunan \u00f6rt\u00fck bir kelime s\u0131n\u0131f\u0131 i\u00e7inde birlikte e\u015fleme bi\u00e7iminde \u00e7al\u0131\u015fmaktad\u0131r. \u00d6\u011frenilen yakla\u015f\u0131m \u00f6rt\u00fck yap\u0131daki her bir veri nokta b\u00fct\u00fcn\u00fcn\u00fcn (bu bir belgedir), tek bir \u00f6rt\u00fck s\u0131n\u0131fa olas\u0131l\u0131\u011fa dayal\u0131 olarak atand\u0131\u011f\u0131 (Collins ve Lanza, 2010), geleneksel \u00f6rt\u00fck s\u0131n\u0131f modellerinden daha karma\u015f\u0131kt\u0131r. GDT modeline, veri noktalar\u0131n\u0131n \u00f6rt\u00fck s\u0131n\u0131flar\u0131n bir bile\u015fimi olarak g\u00f6r\u00fclerek belgelerin i\u00e7indeki s\u00f6zc\u00fcklerin faraz\u00ee olarak \u00f6rt\u00fck s\u0131n\u0131flara atand\u0131\u011f\u0131 ek bir yap\u0131 katman\u0131 daha konmu\u015ftur. Bu konu analizi i\u00e7in \u00f6nemli bir yap\u0131d\u0131r. Belge temsillerinin \u00f6rt\u00fck s\u00f6zc\u00fck s\u0131n\u0131flar\u0131n\u0131n herhangi bir bile\u015fimi olmas\u0131na izin vererek temalar\u0131n bireysel belgelerin i\u00e7inde harmanlanmas\u0131 esnekli\u011finin de korunmas\u0131yla \u00f6rt\u00fck s\u0131n\u0131flar\u0131n say\u0131s\u0131n\u0131n y\u00f6netilebilir bir boyutta kalmas\u0131 m\u00fcmk\u00fcn k\u0131l\u0131nabilir. Her bir \u00f6rt\u00fck s\u00f6zc\u00fck s\u0131n\u0131f\u0131 s\u00f6zc\u00fcklerin bir da\u011f\u0131l\u0131m\u0131yla temsil edilir. Da\u011f\u0131l\u0131mda en y\u00fcksek say\u0131da yer alan s\u00f6zc\u00fckler, ili\u015fkili \u00f6rt\u00fck s\u0131n\u0131f veya konunun en ay\u0131rt edici \u00f6zelli\u011fi olarak ele al\u0131n\u0131rlar.<\/p>\n

GDT’nin denetimsiz dil i\u015fleme tekniklerinden bir olmas\u0131ndan dolay\u0131 belirlenen temalar\u0131n konu temalar\u0131n\u0131n d\u00fczenlemesine dair insan \u00f6nsezisiyle tam olarak e\u015fle\u015fece\u011fini beklemek \u00e7ok da mant\u0131kl\u0131 olmayacakt\u0131r ancak s\u00f6zc\u00fcklerin birlikte bulunma ili\u015fkilerinin modellendi\u011fi bir teknik olarak ili\u015fkilendirilebilece\u011fi d\u00fc\u015f\u00fcn\u00fclen baz\u0131 \u015feylerin belirlenmesini beklemek m\u00fcmk\u00fcnd\u00fcr. GDT \u00f6z\u00fcnde bir veri azaltma tekni\u011fidir. G\u00fc\u00e7l\u00fc oldu\u011fu y\u00f6nler derlemde \u00e7ok yayg\u0131n olan, s\u0131kl\u0131kla ortak temalara kar\u015f\u0131l\u0131k gelen s\u00f6zc\u00fck ili\u015fkilendirmelerinin belirlemesinden kaynaklanmaktad\u0131r. Ancak ortak\/yayg\u0131n temalar\u0131n ilgilenilen temalarla bire bir uyum olmas\u0131 gerekmez. Ne yaz\u0131k ki, bu sonu\u00e7ta ortaya \u00e7\u0131kan g\u00f6sterimde ilgilenilen temalar\u0131n ortak\/yayg\u0131n olmayanlar\u0131 i\u00e7in farkl\u0131 bir g\u00f6sterim olu\u015fturulmayacakt\u0131r. Benzer olarak, yayg\u0131n kullan\u0131lan fikirlerin al\u0131\u015f\u0131lmad\u0131k ifadeleri genel olarak GDT sahas\u0131 i\u00e7inde bir sezgisel g\u00f6sterimle e\u015fle\u015fmeyi ba\u015faramayacakt\u0131r. Metin verisinin g\u00f6sterimi ayr\u0131 bir \u00f6neme sahiptir. S\u0131kl\u0131kla, GDT modelleri bireysel s\u00f6zc\u00fck niteliklerinden olu\u015fturulan nitelik alanlar\u0131 \u00fczerinden hesaplan\u0131r. B\u00f6ylelikle, bireysel s\u00f6zc\u00fcklerle yakalanmayan hi\u00e7bir \u015fey model taraf\u0131ndan eri\u015filebilir olmayacakt\u0131r.<\/p>\n

DENET\u0130ML\u0130 Y\u00d6NTEMLER<\/h2>\n

Spektrumun di\u011fer ucunda ise denetimli y\u00f6ntemler bulunmaktad\u0131r. Biraz fazlaca basitle\u015ftirilmi\u015f bir bak\u0131\u015f a\u00e7\u0131s\u0131yla ele al\u0131n\u0131rsa, denetimli makine \u00f6\u011frenimi y\u00f6ntemleri, genellikle s\u0131n\u0131f de\u011feri olarak adland\u0131r\u0131lan bir sonu\u00e7 \u00f6zelli\u011fi ile nitelik olarak adland\u0131r\u0131lan tahmin unsurlar\u0131ndan olu\u015fan bir koleksiyonu ili\u015fkilendirecek vekt\u00f6r k\u00fcmeleri \u00fczerinde \u00e7al\u0131\u015fan algoritmalard\u0131r. Son zamanlarda, denetimli makine \u00f6\u011frenmesinin \u00f6\u011frenme s\u00fcre\u00e7lerinin de\u011ferlendirilmesi problemine uygulanmas\u0131 tart\u0131\u015fma konudur. Bu problem otomatik- i\u015fbirlikli \u00f6\u011frenme s\u00fcre\u00e7 analizi olarak adland\u0131r\u0131lmaktad\u0131r. \u0130\u015fbirlikli s\u00fcre\u00e7lerin otomatik analizi, i\u015fbirlikli \u00f6\u011frenme s\u0131ras\u0131nda ger\u00e7ek zamanl\u0131 de\u011ferlendirme, i\u015fbirlikli-\u00f6\u011frenme oturumlar\u0131n\u0131n ortas\u0131nda destekleyici m\u00fcdahaleleri dinamik olarak tetikleme ve i\u015fbirlikli-\u00f6\u011frenme s\u00fcre\u00e7lerinin etkili bir \u015fekilde analiz edilmesini kolayla\u015ft\u0131rmak i\u00e7in de\u011fere sahiptir. Bu dinamik yakla\u015f\u0131m\u0131n di\u011fer e\u015fde\u011fer statik destek yakla\u015f\u0131m\u0131ndan daha etkili oldu\u011fu g\u00f6sterilmi\u015ftir Kumar, Ros\u00e9, Wang, Joshi ve Robinson, 2007). Otomatikle\u015ftirilmi\u015f i\u015fbirlikli \u00f6\u011frenme s\u00fcre\u00e7 analizindeki erken d\u00f6nem yap\u0131lm\u0131\u015f \u00e7al\u0131\u015fmalar metin temelli etkile\u015fimleri ve t\u0131klama dizisi verisine odaklanmaktayd\u0131 (Soller ve Lesgold, 2007; Erkens ve Janssen, 2008; Ros\u00e9 vd., 2008; McLaren vd., 2007; Mu vd., 2012). \u0130\u015fbirlikli s\u00fcre\u00e7lerin konu\u015fmayla analizine y\u00f6nelik eski \u00e7al\u0131\u015fmalar da ortaya \u00e7\u0131kmaktad\u0131r (Gweon vd., 2013; Gweon, Agarwal, Udani, Raj ve Ros\u00e9, 2011). Dilbilim ve psikolojideki teorik \u00e7er\u00e7evelerden hareketle olu\u015fturulmu\u015f g\u00f6sterimlerin umut vadetmesi ise tutarl\u0131 bir bulgudur (Ros\u00e9 ve Tovares, bas\u0131m a\u015famas\u0131nda; Wen, Yang ve Ros\u00e9, 2014b; Gweon vd., 2013; Ros\u00e9 ve VanLehn, 2005). Alanda deneyim edinmek i\u00e7in LightSIDE ara\u00e7 tezgah\u0131n\u0131n iyi bir ba\u015flang\u0131\u00e7 noktas\u0131 oldu\u011fundan daha \u00f6nce de bahsetmi\u015ftik.<\/p>\n

B\u0130R ADIM \u00d6TEYE GE\u00c7ERKEN<\/h2>\n

SA alan\u0131 ile ilgili daha fazla bilgi edinmeye istekli okuyucular ilk olarak, temel alan yaz\u0131na biraz dalmakla fayda sa\u011flayacaklard\u0131r. Bu dilbilim alan\u0131, (Levinson, 1983; O’Grady vd., 2009), s\u00f6ylem analizi (Martin ve Rose, 2003; Martin ve White, 2005; Biber ve Conrad, 2011) ve dil teknolojileri (Manning ve Schuetze, 1999; Jurafsky ve Martin, 2009; Jackson ve Moulinier, 2007) alanlar\u0131na dayand\u0131r\u0131lm\u0131\u015ft\u0131r.<\/p>\n

KAYNAK\u00c7A<\/h2>\n

Adamopoulos, P. (2013). What makes a great MOOC? An interdisciplinary analysis of student retention in online courses. Proceedings of the 34th International Conference on Information Systems: Reshaping Society through Information Systems Design <\/i>(ICIS 2013), 15\u201318 December 2013, Milan, Italy. http:\/\/aisel.aisnet.org\/ icis2013\/proceedings\/BreakthroughIdeas\/13\/ <\/span><\/p>\n

Allen, L., Snow, E., & McNamera, D. (2015). Are you reading my mind? Modeling students\u2019 reading comprehension skills with natural language processing techniques. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 246\u2013254). New York: ACM. <\/span><\/p>\n

Biber, D., & Conrad, S. (2011). Register, Genre, and Style<\/i>. Cambridge, UK: Cambridge University Press. <\/span><\/p>\n

Blei, D., Ng, A., & Jordan, M. (2003). Latent Dirichlet allocation. Journal of Machine Learning Research, 3<\/i>, 993\u20131022. <\/span><\/p>\n

Buckingham Shum, S. (2013). Proceedings of the 1st International Workshop on Discourse-Centric Learning Analytics (DCLA13), 8 April 2013, Leuven, Belgium. <\/span><\/p>\n

Buckingham Shum, S., de Laat, M., de Liddo, A., Ferguson, R., & Whitelock, D. (2014). Proceedings of the 2nd International Workshop on Discourse-Centric Learning Analytics <\/i>(DCLA14), 24 March 2014, Indianapolis, IN, USA. <\/span><\/p>\n

Chen, B., Chen, X., & Xing, W. (2015). \u201cTwitter archaeology\u201d of Learning Analytics and Knowledge conferences. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 340\u2013349). New York: ACM. <\/span><\/p>\n

Collins, L., & Lanza, S. T. (2010). Latent class and latent transition analysis with applications in the social, behavioral, and health sciences. Wiley. <\/span><\/p>\n

Dascalu, M., Dessus, P., & McNamera, D. (2015). Discourse cohesion: A signature of collaboration. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 350\u2013354). New York: ACM. <\/span><\/p>\n

Erkens, G., & Janssen, J. (2008). Automatic coding of dialogue acts in collaboration protocols. International Journal of Computer-Supported Collaborative Learning, 3<\/i>, 447\u2013470. <\/span><\/p>\n

Ezen-Can, A., Boyer, K., Kellog, S., & Booth, S. (2015). Unsupervised modeling for understanding MOOC discussion forums: A learning analytics approach. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 146\u2013150). New York: ACM. <\/span><\/p>\n

Foltz, P. (1996). Latent semantic analysis for text-based research. Behavior Research Methods, Instruments, & Computers, 28<\/i>(2), 197\u2013202. <\/span><\/p>\n

Garson, G.D. (2013). Factor Analysis. Asheboro, NC: Statistical Associates Publishing. http:\/\/www.statisticalassociates.com\/factoranalysis.htm <\/span><\/p>\n

Gianfortoni, P., Adamson, D., & Ros\u00e9, C. P. (2011). Modeling stylistic variation in social media with stretchy patterns. Proceedings of the 1st Workshop on Algorithms and Resources for Modeling of Dialects and Language Varieties <\/i>(DIALECTS\u201911), 31 July 2011, Edinburgh, Scotland (pp. 49\u201359). Association for Computational Linguistics. <\/span><\/p>\n

Griffiths, T. L., & Steyvers, M. (2004). Finding scientific topics. Proceedings of the National Academy of Sciences, 101<\/i>, 5228\u20135235. <\/span><\/p>\n

Gweon, G., Jain, M., McDonough, J., Raj, B., & Ros\u00e9, C. P. (2013). Measuring prevalence of other-oriented transactive contributions using an automated measure of speech style accommodation. International Journal of Computer Supported Collaborative Learning, 8<\/i>(2), 245\u2013265. <\/span><\/p>\n

Gweon, G., Agarwal, P., Udani, M., Raj, B., & Ros\u00e9, C. P. (2011). The automatic assessment of knowledge integration processes in project teams. Proceedings of the 9th International Conference on Computer-Supported Collaborative Learning, Volume 1: Long Papers <\/i>(CSCL 2011), 4\u20138 July 2011, Hong Kong, China (pp. 462\u2013469). International Society of the Learning Sciences.<\/span><\/p>\n

Hmelo-Silver, C., Chinn, C., Chan, C., & O\u2019Donnell, A. (2013). The International Handbook of Collaborative Learning<\/i>. Routledge. <\/span><\/p>\n

Hsiao, I., & Awasthi, P. (2015). Topic facet modeling: Semantic and visual analytics for online discussion forums. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 231\u2013235). New York: ACM. <\/span><\/p>\n

Jackson, P., & Moulinier, I. (2007). Natural language processing for online applications: Text retrieval, extraction, and categorization. Amsterdam: John Benjamins Publishing Company. <\/span><\/p>\n

Jo, Y., Loghmanpour, N., & Ros\u00e9, C. P. (2015). Time series analysis of nursing notes for mortality prediction via state transition topic models. Proceedings of the 24th ACM International Conference on Information and Knowledge Management <\/i>(CIKM\u201915), 19\u201323 October 2015, Melbourne, VIC, Australia (pp. 1171\u20131180). New York: ACM. <\/span><\/p>\n

Joksimovi\u0107, S., Kovanovi\u0107, V., Jovanovi\u0107, J., Zouaq, A., Ga\u0161evi\u0107, D., & Hatala, M. (2015). What do cMOOC participants talk about in social media? A topic analysis of discourse in a cMOOC. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 156\u2013165). New York: ACM. <\/span><\/p>\n

Jurafsky, D., & Martin, J. (2009). Speech and language processing: An introduction to natural language processing, computational linguistics, and speech recognition. Pearson. <\/span><\/p>\n

Knight, S., & Littleton, K. (2015). Developing a multiple-document-processing performance assessment for epistemic literacy. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 241\u2013245). New York: ACM. <\/span><\/p>\n

Kumar, R., Ros\u00e9, C. P., Wang, Y. C., Joshi, M., & Robinson, A. (2007). Tutorial dialogue as adaptive collaborative learning support. Proceedings of the 13th International Conference on Artificial Intelligence in Education: Building Technology Rich Learning Contexts that Work <\/i>(AIED 2007), 9\u201313 July 2007, Los Angeles, CA, USA (pp. 383\u2013390). IOS Press. <\/span><\/p>\n

Levinson, S. (1983). Conversational structure. Pragmatics <\/i>(pp. 284\u2013286). Cambridge, UK: Cambridge University Press. <\/span><\/p>\n

Loehlin, J.C. (2004). Latent variable models: An introduction to factor, path, and structural equation analysis. Routledge. <\/span><\/p>\n

Manning, C., & Schuetze, H. (1999). Foundations of statistical natural language processing. MIT Press. <\/span><\/p>\n

Martin, J., & Rose, D. (2003). Working with discourse: Meaning beyond the clause. Continuum. <\/span><\/p>\n

Martin, J., & White, P. (2005). The language of evaluation: Appraisal in English. Palgrave. <\/span><\/p>\n

Mayfield, E., & Ros\u00e9, C. P. (2013). LightSIDE: Open source machine learning for text accessible to non-experts. In M. D. Shermis & J. Burstein (Eds.), Handbook of Automated Essay Grading: Current Applications and New Directions <\/i>(pp. 124\u2013135). Routledge. <\/span><\/p>\n

McLaren, B., Scheuer, O., De Laat, M., Hever, R., de Groot, R., & Ros\u00e9, C. P. (2007). Using machine learning techniques to analyze and support mediation of student E-discussions. Proceedings of the 13th International Conference on Artificial Intelligence in Education: Building Technology Rich Learning Contexts That Work <\/i>(AIED 2007), 9\u201313 July 2007, Los Angeles, CA, USA (pp. 331\u2013338). IOS Press. <\/span><\/p>\n

McNamara, D. S., & Graesser, A. C. (2012). Coh-Metrix: An automated tool for theoretical and applied natural language processing. In P. M. McCarthy & C. Boonthum (Eds.), Applied Natural Language Processing: Identification, Investigation, and Resolution <\/i>(pp. 188\u2013205). Hershey, PA: IGI Global. <\/span><\/p>\n

Milligan, S. (2015). Crowd-sourced learning in MOOCs: Learning analytics meets measurement theory. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 151\u2013155). New York: ACM. <\/span><\/p>\n

Molenaar, I., & Chiu, M. (2015). Effects of sequences of socially regulated learning on group performance. Proceedings of the 5th Internation Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 236\u2013240). New York: ACM.<\/span><\/p>\n

Morrow, R. A., & Brown, D. D. (1994). Deconstructing the conventional discourse of methodology: Quantitative versus qualitative methods. In R. A. Morrow & D. D. Brown (Eds.), Critical theory and methodology: Contemporary social theory<\/i>, Vol. 3 (pp. 199\u2013225). Thousand Oaks, CA: Sage. <\/span><\/p>\n

Mu, J., Stegmann, K., Mayfield, E., Ros\u00e9, C. P., & Fischer, F. (2012). The ACODEA framework: Developing segmentation and classification schemes for fully automatic analysis of online discussions. International Journal of Computer-Supported Collaborative Learning, 138<\/i>, 285\u2013305. <\/span><\/p>\n

Nehm, R., Ha, M., & Mayfeld, E. (2012). Transforming biology assessment with machine learning: Automated scoring of written evolutionary explanations. Journal of Science Education and Technology, 21<\/i>, 183\u2013196. <\/span><\/p>\n

Nguyen, D., Dogru\u00f6z, A. S., Ros\u00e9, C. P., & de Jong, F. (in press). Computational sociolinguistics: A survey. Computational Linguistics<\/i>. <\/span><\/p>\n

O\u2019Donnell, A., & King, A. (1999). Cognitive perspectives on peer learning<\/i>. Routledge. <\/span><\/p>\n

O\u2019Grady, W., Archibald, J., Aronoff, M., & Rees-Miller, J. (2009). Contemporary linguistics: An introduction<\/i>. Boston \/ New York: Bedford \/ St. Martin. <\/span><\/p>\n

Page, E. B. (1966). The imminence of grading essays by computer. Phi Delta Kappan, 48<\/i>, 238\u2013243. <\/span><\/p>\n

Pang, B., & Lee, L. (2008). Opinion mining and sentiment analysis. Foundations and Trends in Information Retrieval, 2<\/i>(1\u20132), 1\u2013135. <\/span><\/p>\n

Ramesh, A., Goldwasser, D., Huang, B., Daum\u00e9 III, H., & Getoor, L. (2013). Modeling learner engagement in MOOCs using probabilistic soft logic. NIPS Workshop on Data Driven Education: Advances in Neural Information Processing Systems <\/i>(NIPS-DDE 2013), 9 December 2013, Lake Tahoe, NV, USA. https: \/\/www.umiacs. umd.edu\/~hal\/docs\/daume13engagementmooc.pdf <\/span><\/p>\n

Ribeiro, B.T. (2006). Footing, positioning, voice: Are we talking about the same things? In A. De Fina, D. Schiffrin, & M. Bamberg (Eds.), Discourse and identity <\/i>(pp. 48\u201382). New York: Cambridge University Press. <\/span><\/p>\n

Ros\u00e9, C. P., & Tovares, A. (2015). What sociolinguistics and machine learning have to say to one another about interaction analysis. In L. Resnick, C. Asterhan, & S. Clarke (Eds.), Socializing intelligence through academic talk and dialogue<\/i>. Washington, DC: American Educational Research Association. <\/span><\/p>\n

Ros\u00e9, C. P., & VanLehn, K. (2005). An evaluation of a hybrid language understanding approach for robust selection of tutoring goals. International Journal of Artificial Intelligence in Education, 15<\/i>, 325\u2013355. <\/span><\/p>\n

Ros\u00e9, C. P., Wang, Y. C., Cui, Y., Arguello, J., Stegmann, K., Weinberger, A., & Fischer, F., (2008). Analyzing collaborative learning processes automatically: Exploiting the advances of computational linguistics in computer-supported collaborative learning. International Journal of Computer Supported Collaborative Learning, 3(3), 237\u2013271. <\/span><\/p>\n

Sekiya, T., Marsuda, Y., & Yamaguchi, K. (2015). Curriculum analysis of CS departments based on CS2013 by simplified, supervised LDA. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 330\u2013339). New York: ACM. <\/span><\/p>\n

Shermis, M. D., & Burstein, J. (2013). Handbook of Automated Essay Evaluation: Current Applications and New Directions. New York: Routledge. <\/span><\/p>\n

Shermis, M., & Hammer, B. (2012). Contrasting state-of-the-art automated scoring of essays: Analysis. Annual National Council on Measurement in Education Meeting, 14\u201316. <\/span><\/p>\n

Simsek, D., Sandor, A., & Buckingham Shum, S. (2015). Correlations between automated rhetorical analysis and tutor\u2019s grades on student essays. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 355\u2013359). New York: ACM. <\/span><\/p>\n

Skrondal, A., & Rabe-Hesketh, S. (2004). Generalized latent variable modeling: Multi-level, longitudinal, and structural equation models. Chapman & Hall\/CRC. <\/span><\/p>\n

Snow, E., Allen, L., Jacovina, M., Perret, C., & McNamera, D. (2015). You\u2019ve got style: Writing flexibility across time. Proceedings of the 5th International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201915), 16\u201320 March 2015, Poughkeepsie, NY, USA (pp. 194\u2013202). New York: ACM.<\/span><\/p>\n

Soller, A., & Lesgold, A. (2007). Modeling the process of collaborative learning. In H. U. Hoppe, H. Ogata, & A. Soller (Eds.), The role of technology in CSCL: Studies in technology enhanced collaborative learning <\/i>(pp 63\u201386). Springer. doi:10.1007\/978-0-387-71136-2_5 <\/span><\/p>\n

Wen, M., Yang, D., & Ros\u00e9, C. P. (2014a). Sentiment analysis in MOOC discussion forums: What does it tell us? In J. Stamper, Z. Pardos, M. Mavrikis, & B. M. McLaren (Eds.), Proceedings of the 7th International Conference on Educational Data Mining <\/i>(EDM2014), 4\u20137 July, London, UK. International Educational Data Mining Society. https:\/\/www.researchgate.net\/publication\/264080975_Sentiment_analysis_in_MOOC_discussion_forums_What_does_it_tell_us <\/span><\/p>\n

Wen, M., Yang, D., & Ros\u00e9, C. P. (2014b). Linguistic reflections of student engagement in massive open online courses. Proceedings of the 8th International AAAI Conference on Weblogs and Social Media (ICWSM\u201914), 1\u20134 June 2014, Ann Arbor, Michigan, USA. Palo Alto, CA: AAAI Press. http:\/\/www.cs.cmu.edu\/~mwen\/papers\/ icwsm2014-camera-ready.pdf <\/span><\/p>\n

Witten, I. H., Frank, E., & Hall, M. (2011). Data mining: Practical machine learning tools and techniques<\/i>, 3rd ed. San Francisco, CA: Elsevier.<\/span><\/p>\n

\n
\n

1<\/a> https:\/\/www.edx.org\/course\/data\u2013analytics\u2013learning\u2013utarlingtonx\u2013link5\u201310x<\/span><\/p>\n<\/div>\n

\n

2<\/a> http:\/\/lightsidelabs.com\/research\/<\/span><\/p>\n<\/div>\n

\n

3<\/a> orj. affect analysis<\/span><\/p>\n<\/div>\n

\n

4<\/a> orj. sentiment analysis <\/span><\/p>\n<\/div>\n

\n

5<\/a> orj. attrition<\/span><\/p>\n<\/div>\n","protected":false},"author":1,"menu_order":5,"template":"","meta":{"pb_show_title":"on","pb_short_title":"B\u00f6l\u00fcm 9 Do\u011fal Dil \u0130\u015fleme ve \u00d6\u011frenme Analiti\u011fi","pb_subtitle":"","pb_authors":[],"pb_section_license":""},"chapter-type":[48],"contributor":[],"license":[],"class_list":["post-53","chapter","type-chapter","status-publish","hentry","chapter-type-numberless"],"part":46,"_links":{"self":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/chapters\/53","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/chapters"}],"about":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/wp\/v2\/types\/chapter"}],"author":[{"embeddable":true,"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/wp\/v2\/users\/1"}],"version-history":[{"count":0,"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/chapters\/53\/revisions"}],"part":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/parts\/46"}],"metadata":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/chapters\/53\/metadata\/"}],"wp:attachment":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/wp\/v2\/media?parent=53"}],"wp:term":[{"taxonomy":"chapter-type","embeddable":true,"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/chapter-type?post=53"},{"taxonomy":"contributor","embeddable":true,"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/wp\/v2\/contributor?post=53"},{"taxonomy":"license","embeddable":true,"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/wp\/v2\/license?post=53"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}