{"id":104,"date":"2020-09-03T16:39:46","date_gmt":"2020-09-03T13:39:46","guid":{"rendered":"http:\/\/acikkitap.com.tr\/oaek\/chapter\/bolum-20-ogrenme-analitigi-icin-tavsiye-sistemlerini-uygulamak-akilli-ogreticiler\/"},"modified":"2020-09-03T16:39:46","modified_gmt":"2020-09-03T13:39:46","slug":"bolum-20-ogrenme-analitigi-icin-tavsiye-sistemlerini-uygulamak-akilli-ogreticiler","status":"publish","type":"chapter","link":"https:\/\/acikkitap.com.tr\/oaek\/chapter\/bolum-20-ogrenme-analitigi-icin-tavsiye-sistemlerini-uygulamak-akilli-ogreticiler\/","title":{"raw":"B\u00f6l\u00fcm 20 \u00d6\u011frenme Analiti\u011fi \u0130\u00e7in Tavsiye Sistemlerini Uygulamak: Ak\u0131ll\u0131 \u00d6\u011freticiler","rendered":"B\u00f6l\u00fcm 20 \u00d6\u011frenme Analiti\u011fi \u0130\u00e7in Tavsiye Sistemlerini Uygulamak: Ak\u0131ll\u0131 \u00d6\u011freticiler"},"content":{"raw":"\n

Soude Fazeli1<\/sup>, Hendrik Drachsler2<\/sup>, Peter Sloep2<\/sup><\/span><\/span><\/p>\n

1<\/sup>Welten Enstit\u00fcs\u00fc, \u00d6\u011frenme, \u00d6\u011fretme ve Teknoloji Ara\u015ft\u0131rma Merkezi, Hollanda<\/span><\/span><\/p>\n

2<\/sup>Hollanda A\u00e7\u0131k \u00dcniversitesi, Hollanda<\/span><\/span><\/p>\n

DOI: 10.18608\/hla17.020<\/span><\/span><\/p>\n\n

\u00d6Z<\/h2>\n

Bu b\u00f6l\u00fcm, bir \u00f6neri sistem deneyiminin \u00f6\u011frenme analiti\u011fi (\u00d6A) alan\u0131nda nas\u0131l uygulanabilece\u011fine ili\u015fkin bir \u00f6rnek sunmaktad\u0131r. Bu b\u00f6l\u00fcmde sunulan \u00f6rnek \u00e7al\u0131\u015fma, \u00f6\u011frenmede \u00f6neri sistemlerini de\u011ferlendirmek i\u00e7in standart bir y\u00f6ntem izlemi\u015ftir. \u00d6rnek, Avrupa'daki e\u011fitim payda\u015flar\u0131 i\u00e7in, Facebook benzeri bir sosyal \u00f6\u011frenme platformu sa\u011flamay\u0131 ancak Facebook' dan farkl\u0131 olarak sadece \u00f6\u011frenme ve bilgi payla\u015f\u0131m\u0131n\u0131 ama\u00e7layan, FP7 programlar\u0131nda A\u00e7\u0131k Ke\u015fif Alan\u0131 (AKA) projesi kapsam\u0131nda haz\u0131rlanm\u0131\u015ft\u0131r. Bu b\u00f6l\u00fcmde, ad\u0131m ad\u0131m bir s\u00fcre\u00e7te tam bir tavsiye sistemi veri \u00e7al\u0131\u015fmas\u0131n\u0131 a\u00e7\u0131klamaktay\u0131z. Ayr\u0131ca, \u00f6\u011frenme alan\u0131ndaki veri g\u00fcd\u00fcml\u00fc \u00e7al\u0131\u015fmalar\u0131n eksikliklerini ana hatlar\u0131yla belirtmekte ve SoLAR toplulu\u011fu taraf\u0131ndan \u00f6nerildi\u011fi gibi a\u00e7\u0131k bir \u00f6\u011frenme analiti\u011fi platformuna olan ihtiyac\u0131 vurgulamaktay\u0131z.<\/span><\/span><\/p>\n

Anahtar Kelimeler<\/span>: Tavsiye sistemi, \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmas\u0131, kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131, i\u015fbirlikli filtreleme, bilgi arama ve alma, bilgi filtreleme, ara\u015ft\u0131rma, metodoloji, seyreklik <\/span><\/span><\/p>\n

\u00c7e\u015fitli alanlarda b\u00fcy\u00fck miktarlarda verinin ortaya \u00e7\u0131kmas\u0131yla birlikte, tavsiye sistemleri, kullan\u0131c\u0131lar\u0131n ge\u00e7mi\u015f davran\u0131\u015flar\u0131na ve mevcut durumlar\u0131na g\u00f6re en uygun bilgileri sa\u011flamak i\u00e7in pratik bir yakla\u015f\u0131m haline gelmi\u015ftir. Duval (2011) tavsiye sistemlerini \"'Se\u00e7im paradoksunu' ele almak ve bollu\u011fu bir problemden \u00f6\u011frenme i\u00e7in bir varl\u0131\u011fa d\u00f6n\u00fc\u015ft\u00fcrmek\u201din bir \u00e7\u00f6z\u00fcm\u00fc olarak ortaya koymu\u015ftur (s. 9), e\u011fitsel veri madencili\u011fi, b\u00fcy\u00fck veri ve web analiti\u011fi gibi alanlar\u0131n t\u00fcm\u00fc b\u00fcy\u00fck miktarda veride \u00f6r\u00fcnt\u00fc bulmaya \u00e7al\u0131\u015ft\u0131\u011f\u0131na dikkat \u00e7ekiyor. \u00d6rne\u011fin, veri madencili\u011fi yakla\u015f\u0131mlar\u0131, kullan\u0131c\u0131lar\u0131n toplanan verilerinden tespit edilen benzerlik \u00f6r\u00fcnt\u00fclerine dayanarak tavsiyelerde bulunabilir. Bunun yan\u0131 s\u0131ra, Greller ve Drachsler (2012) taraf\u0131ndan yap\u0131lan bir saha ara\u015ft\u0131rmas\u0131, tavsiye sistemlerini ve ki\u015fiselle\u015ftirmeyi \u00d6A ara\u015ft\u0131rmas\u0131n\u0131n \u00f6nemli bir par\u00e7as\u0131 olarak tan\u0131mlam\u0131\u015ft\u0131r.<\/span><\/p>\n

Tavsiye sistemleri teknolojilerinin temeline ve algoritmalar\u0131na g\u00f6re ay\u0131rt edilebilir. Kabaca i\u00e7erik tabanl\u0131 ya da i\u015fbirlikli filtreleme kullan\u0131rlar. \u0130\u00e7erik tabanl\u0131 algoritmalar, tavsiye sistemlerinde kullan\u0131lan ana y\u00f6ntemlerden biridir; \u00f6genin i\u00e7eri\u011finin temsilini kullan\u0131c\u0131n\u0131n tercih modeliyle kar\u015f\u0131la\u015ft\u0131rarak kullan\u0131c\u0131ya bir \u00f6ge tavsiye ederler (Pazzani ve Billsus, 2007). Ortak filtreleme, kullan\u0131c\u0131lar\u0131n \u00f6geleri hakk\u0131ndaki g\u00f6r\u00fc\u015flerine ve geri bildirimlerine dayan\u0131r. \u0130\u015fbirlikli filtreleme algoritmalar\u0131 ilk \u00f6nce benzer d\u00fc\u015f\u00fcnen kullan\u0131c\u0131lar\u0131 bulur ve bunlar\u0131 baz\u0131 hedef kullan\u0131c\u0131lara en yak\u0131n kom\u015fular olarak tan\u0131t\u0131r; daha sonra, bu kullan\u0131c\u0131n\u0131n hedeflenen kullan\u0131c\u0131lar\u0131n en yak\u0131n kom\u015fular\u0131 (e\u015f dereceleri) taraf\u0131ndan verilen \u00f6gelere g\u00f6re bir \u00f6genin derecelendirmesini tahmin ederler (Herlocker, Konstan, Terveen ve Riedl, 2004; Manouselis, Drachsler, Verbert ve Duval, 2012; Schafer, Frankowski, Herlocker ve Sen, 2007).<\/span><\/p>\n

Ge\u00e7mi\u015fte, farkl\u0131 ama\u00e7larla \u00e7e\u015fitli e\u011fitim projelerinde tavsiye sistemlerini uygulad\u0131k (Drachsler vd., 2010; Fazeli, Loni, Drachsler ve Sloep, 2014; Drachsler vd., 2009). Bu b\u00f6l\u00fcmde, e\u011fitimde tavsiye sistemi algoritmalar\u0131n\u0131n geli\u015ftirilmesi ve de\u011ferlendirilmesi ile ilgili \u015fu ana kadar tespit etti\u011fimiz en iyi uygulamalardan baz\u0131lar\u0131n\u0131 payla\u015fmak istiyoruz; \u00f6zellikle de bir tavsiye sistemleri denemesinin nas\u0131l kurulaca\u011f\u0131 ve \u00e7al\u0131\u015ft\u0131r\u0131laca\u011f\u0131na dair bir \u00f6rnek vermek istiyoruz.<\/span><\/p>\n

Teknoloji ile Geli\u015ftirilmi\u015f \u00d6\u011frenme alan\u0131na istinaden TavSisTD\u00d6 \u00e7al\u0131\u015fma grubunun Tavsiye Sistemleri i\u00e7in yapt\u0131\u011f\u0131 a\u00e7\u0131klamaya g\u00f6re (Drachsler, Verbert, Santos ve Manouselis) standart bir de\u011ferlendirme y\u00f6nteminin uygulanmas\u0131 \u00f6nemlidir. \u00c7al\u0131\u015fma grubu, e\u011fitimde bir tavsiye sistemini de\u011ferlendirmek i\u00e7in d\u00f6rt kritik ad\u0131mdan olu\u015fan bir ara\u015ft\u0131rma metodolojisi tan\u0131mlam\u0131\u015ft\u0131r:<\/span><\/p>\n\n

    \n \t
  1. \n

    Tavsiye g\u00f6revine uyan bir dizi<\/i><\/span> veri k\u00fcmesinin se\u00e7imi. \u00d6rne\u011fin, tavsiye g\u00f6revi bir kullan\u0131c\u0131 i\u00e7in yeni \u00f6geler bulmak veya ilgili \u00f6geleri ke\u015ffetmek olabilir.<\/span><\/p>\n<\/li>\n \t

  2. \n

    Tavsiye edilen sistemlerin performans\u0131 hakk\u0131nda bilgi vermek i\u00e7in iyi bilinen veri k\u00fcmeleri (m\u00fcmk\u00fcnse<\/i><\/span>, MovieLens gibi e\u011fitime dayal\u0131 veri k\u00fcmeleri d\u00e2hil) d\u00e2hil olmak \u00fczere se\u00e7ilen veri k\u00fcmelerinde farkl\u0131 algoritmalar\u0131n \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmas\u0131.<\/span><\/p>\n<\/li>\n \t

  3. \n

    Tasarlanan tavsiye sisteminin teknik y\u00f6nlerinin yan\u0131 s\u0131ra \u00f6\u011frenenler \u00fczerindeki psiko-e\u011fitsel etkilerini test etmek i\u00e7in kapsaml\u0131 bir kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131.<\/span><\/p>\n<\/li>\n \t

  4. \n

    Tavsiye sisteminin ger\u00e7ek kullan\u0131c\u0131lar ile ger\u00e7ek\u00e7i, normal \u00e7al\u0131\u015fma ko\u015fullar\u0131 alt\u0131nda test edilebilen ger\u00e7ek zamanl\u0131 bir uygulamada kullan\u0131lmas\u0131.<\/span><\/p>\n<\/li>\n<\/ol>\n

    Tavsiye sisteminin tam bir a\u00e7\u0131klamas\u0131 yukar\u0131daki d\u00f6rt ad\u0131ma sunulan s\u0131n\u0131fland\u0131rma \u00e7er\u00e7evesine g\u00f6re eklenmelidir. (Drachsler vd., 2015). Kullan\u0131lan veri k\u00fcmesi, \u00d6\u011frenme Analiti\u011fi Dergisi1<\/a><\/sup>'nin e\u011fitsel veri k\u00fcmeleri ile ilgili \u00f6zel b\u00f6l\u00fcm\u00fcnde bildirilmelidir ve belirli ko\u015fullar alt\u0131nda di\u011fer ara\u015ft\u0131rmac\u0131lar i\u00e7in haz\u0131r bulundurulmal\u0131d\u0131r (Dietze, Siemens, Taibi ve Drachsler, 2016). Bu di\u011fer ara\u015ft\u0131rmac\u0131lar\u0131n kar\u015f\u0131la\u015ft\u0131r\u0131labilir sonu\u00e7lar\u0131 ve yeni g\u00f6r\u00fc\u015fleri elde etmek i\u00e7in ara\u015ft\u0131rman\u0131n herhangi bir b\u00f6l\u00fcm\u00fcn\u00fc tekrarlamas\u0131n\u0131 ve d\u00fczenlemesini ve b\u00f6ylelikle analitik \u00f6\u011frenmenin tavsiye sistemleri gibi bir bilgi birikiminin olu\u015fmas\u0131n\u0131 m\u00fcmk\u00fcn k\u0131lar.<\/span><\/p>\n

    Bu b\u00f6l\u00fcmde e\u011fitimde tavsiye sistemleri i\u00e7in yukar\u0131da a\u00e7\u0131klanan ara\u015ft\u0131rma metodolojisini takip eden deneysel bir \u00e7al\u0131\u015fma \u00f6rne\u011fi sunuyoruz. B\u00f6l\u00fcm\u00fcn geri kalan\u0131 \u015fu \u015fekilde d\u00fczenlenmi\u015ftir: Bu b\u00f6l\u00fcmde, e\u011fitimde tavsiye veren sistemler i\u00e7in yukar\u0131da a\u00e7\u0131klanan ara\u015ft\u0131rma metodolojisini izleyen deneysel bir \u00e7al\u0131\u015fma \u00f6rne\u011fi sunaca\u011f\u0131z. Daha sonra, deneyin pratik sonu\u00e7lar\u0131n\u0131 a\u00e7\u0131klayacak ve sonu\u00e7land\u0131raca\u011f\u0131z.<\/span><\/p>\n\n

    E\u011e\u0130T\u0130M ALANINDA B\u0130R \u00d6NER\u0130 S\u0130STEM\u0130 DENEY\u0130M\u0130<\/h2>\n

    Bu b\u00f6l\u00fcmde, 2014 TD\u00d6AK \u00e7al\u0131\u015fmam\u0131zda sunulan deneysel bir \u00e7al\u0131\u015fmay\u0131 kullanarak \u00f6\u011frenmede bir tavsiye sisteminin nas\u0131l de\u011ferlendirilmesi gerekti\u011fi a\u00e7\u0131klanmaktad\u0131r (Fazeli vd., 2014). Bu \u00e7al\u0131\u015fma, yukar\u0131da a\u00e7\u0131klanan standart metodolojiyi izler. Bununla birlikte, bu metodolojiye ek bir ad\u0131m daha ekledik: TavSisTD\u00d6'n\u00fcn \u00f6zel bir say\u0131s\u0131nda sunulan (Manouselis vd., 2012), kavramsal \/ teorik bir model geli\u015ftirme (Fazeli vd., 2013).<\/span><\/p>\n

    \u00c7al\u0131\u015fmam\u0131zda hedef \u00e7evremiz genel olarak sosyal \u00f6\u011frenme platformlar\u0131d\u0131r. Sosyal \u00f6\u011frenme platformlar\u0131, Facebook gibi sosyal a\u011flara benzer \u015fekilde \u00e7al\u0131\u015f\u0131r ancak Facebook'un aksine yaln\u0131zca \u00f6\u011frenme ve bilgi payla\u015f\u0131m\u0131 amac\u0131yla geli\u015ftirilmi\u015ftir. Bu nedenle; genellikle \u00f6\u011fretmenler, \u00f6\u011frenciler, \u00f6\u011frenenler, politika yap\u0131c\u0131lar gibi e\u011fitim payda\u015flar\u0131 i\u00e7in ortak bir alan olarak hizmet sunarlar. Hedef sosyal \u00f6\u011frenme platformumuz A\u00e7\u0131k Ke\u015fif Alan\u0131 (AKA)'d\u0131r.2<\/a><\/sup> AKA ana sayfas\u0131nda belirtildi\u011fi gibi,<\/span><\/p>\n

    AKA, e-\u00f6\u011frenme ortam\u0131n\u0131n Avrupa ba\u011flam\u0131nda kar\u015f\u0131la\u015ft\u0131\u011f\u0131 \u00e7e\u015fitli zorluklar\u0131 ele almaktad\u0131r. Aray\u00fcz \u00f6\u011frenciler, \u00f6\u011fretmenler, veliler ve politika yap\u0131c\u0131lar g\u00f6z\u00f6n\u00fcnde bulundurularak tasarlanm\u0131\u015ft\u0131r. AKA \u00fc\u00e7 ana hedefi yerine getirecektir. \u00d6ncelikle, payda\u015flar\u0131, da\u011f\u0131n\u0131k e\u011fitsel veri ambarlar\u0131ndan gelen e-\u00f6\u011frenme kaynaklar\u0131 i\u00e7in tek, entegre bir eri\u015fim noktas\u0131 arac\u0131l\u0131\u011f\u0131yla g\u00fc\u00e7lendirecektir. \u0130kincisi, sosyal a\u011f tarz\u0131 \u00e7ok dilli bir portal kullanarak, e-\u00f6\u011frenme kaynaklar\u0131n\u0131 ve ayn\u0131 zamanda e\u011fitim faaliyetlerinin \u00fcretimi i\u00e7in hizmetler sunarak, anlaml\u0131 e\u011fitim faaliyetlerinin \u00fcretiminde payda\u015flar\u0131 hedef almaktad\u0131r. \u00dc\u00e7\u00fcnc\u00fcs\u00fc, payda\u015flar\u0131n okul e\u011fitiminde benimsemeleri i\u00e7in prototip g\u00f6revi g\u00f6rebilecek yeni e\u011fitim faaliyetlerinin etkisini de\u011ferlendirecektir.<\/span><\/p>\n

    \u00c7al\u0131\u015fmam\u0131z\u0131n temel amac\u0131, hangi tavsiye sisteminin bir sosyal \u00f6\u011frenme platformunun veri ve bilgi ihtiya\u00e7lar\u0131na en iyi \u015fekilde cevap verebilece\u011fini bulmakt\u0131r, as\u0131l tavsiye g\u00f6revi kullan\u0131c\u0131lar i\u00e7in ilgili \u00f6geleri bulmakt\u0131r. A\u015fa\u011f\u0131daki alt b\u00f6l\u00fcmlerde, \u00e7al\u0131\u015fmayla ilgili detaylar ad\u0131m ad\u0131m verilmektedir.<\/span><\/p>\n\n

    Veri K\u00fcmesi Se\u00e7imi<\/h3>\n

    \u00c7o\u011fu veri \u00e7al\u0131\u015fmas\u0131 belirli bir ortam\u0131 veya belirli bir kullan\u0131c\u0131 grubunu hedefler ve bu nedenle belirli bir veri t\u00fcr\u00fcn\u00fc gerektirir. Bizim durumumuzda, hedef sosyal \u00f6\u011frenme platformlar\u0131 AKA'd\u0131r. Sonu\u00e7 olarak, AKA'ya benzer \u00f6\u011frenme platformlar\u0131ndan toplanan verileri bulmaya \u00e7al\u0131\u015ft\u0131k. AM\u0130\u00dcV ve OpenScout veri k\u00fcmelerini a\u015fa\u011f\u0131daki nedenlerden dolay\u0131 se\u00e7tik:<\/span><\/p>\n\n

      \n \t
    1. \n

      Veri k\u00fcmeleri, kullan\u0131c\u0131lar\u0131n \u00f6\u011frenme kaynaklar\u0131 hakk\u0131ndaki sosyal verilerini (derecelendirmeler, etiketler, incelemeler, vb.) sa\u011flar. Bu nedenle, veri k\u00fcmelerinin yap\u0131s\u0131, i\u00e7eri\u011fi ve hedef kullan\u0131c\u0131lar\u0131 AKA'n\u0131nkine benzerdir.<\/span><\/p>\n<\/li>\n \t

    2. \n

      Bu veri k\u00fcmeleri \u00fczerinde \u00f6neri sunan algoritmalar\u0131 kullanmak, AKA'n\u0131n ger\u00e7ek kullan\u0131c\u0131lar\u0131yla \u00e7evrimi\u00e7i olmadan \u00f6nce performanslar\u0131n\u0131 de\u011ferlendirmemize yard\u0131mc\u0131 olur.<\/span><\/p>\n<\/li>\n \t

    3. \n

      Hem AM\u0130\u00dcV hem de OpenScout veri k\u00fcmeleri, sosyal verilerin toplanmas\u0131 ve depolanmas\u0131 i\u00e7in standart bir \u00fcst veri belirtimi sunan BO\u00dcV (Ba\u011flamsal Otomatikle\u015ftirilmi\u015f \u00dcst Veri) format\u0131na (Schmitz vd., 2009) uygundur. BO\u00dcV ayr\u0131ca sosyal verilerin depolanmas\u0131 i\u00e7in AKA'da uygulanm\u0131\u015ft\u0131r.<\/span><\/p>\n<\/li>\n<\/ol>\n

      <\/a><\/a><\/a><\/a><\/a> Tablo 20.1. Se\u00e7ilen Veri K\u00fcmelerinin Detaylar\u0131<\/i><\/span><\/span><\/p>\n\n<\/colgroup>\n\n\n\n\n\n
      \n

      Veri k\u00fcmesi<\/span><\/p>\n<\/td>\n

      		\n

      Kullan\u0131c\u0131 say\u0131s\u0131<\/span><\/p>\n<\/td>\n

      		\n

      \u00d6ge say\u0131s\u0131<\/span><\/p>\n<\/td>\n

      		\n

      \u0130\u015flemler<\/span><\/p>\n<\/td>\n

      		\n

      Seyreklik (%)<\/span><\/p>\n<\/td>\n

      		\n

      Kaynak<\/span><\/p>\n<\/td>\n<\/tr>\n

      AM\u0130\u00dcV<\/span><\/span><\/td>\n631<\/span><\/span><\/td>\n12.571<\/span><\/span><\/td>\n23.032<\/span><\/span><\/td>\n99.70<\/span><\/span><\/td>\nAM\u0130\u00dcV portal\u0131<\/span><\/span><\/td>\n<\/tr>\n
      OpenScout<\/span><\/span><\/td>\n331<\/span><\/span><\/td>\n1.568<\/span><\/span><\/td>\n2.560<\/span><\/span><\/td>\n99.50<\/span><\/span><\/td>\nOpenScout portal\u0131<\/span><\/span><\/td>\n<\/tr>\n
      MovieLens<\/span><\/span><\/td>\n941<\/span><\/span><\/td>\n1.512<\/span><\/span><\/td>\n96.719<\/span><\/span><\/td>\n93.69<\/span><\/span><\/td>\nGroupLens ara\u015ft\u0131rmas\u0131<\/span><\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

      Bu iki veri k\u00fcmesinin yan\u0131 s\u0131ra, MovieLens veri k\u00fcmesini referans olarak test ettik, \u00e7\u00fcnk\u00fc bug\u00fcne kadar, e\u011fitim alan\u0131 genel olarak tavsiye sistemleri ile ilgilenen B\u0130MK TavSis konferans serisinin aksine, \u00e7al\u0131\u015fma i\u00e7in referans veri k\u00fcmeleri yetersiz kalm\u0131\u015ft\u0131r. Tablo 20.1, her \u00fc\u00e7 veri k\u00fcmesine genel bir bak\u0131\u015f sunmaktad\u0131r (Fazeli vd., 2014). E\u011fitsel veri k\u00fcmelerinden AM\u0130\u00dcV ve OpenScout\u2019un a\u015f\u0131r\u0131 derecede seyreklikten s\u0131k\u0131nt\u0131 ya\u015fad\u0131\u011f\u0131 dikkate al\u0131nmal\u0131d\u0131r. T\u00fcm verilere ait detaylar TD\u00d6AK 2014 makalemizde daha ayr\u0131nt\u0131l\u0131 olarak tan\u0131mlanmaktad\u0131r (Fazeli vd., 2014).<\/span><\/p>\n\n

      \u00c7evrimd\u0131\u015f\u0131 Veri \u00c7al\u0131\u015fmas\u0131<\/span><\/h2>\n

      Algoritmalar<\/span>. Bu ikinci ad\u0131mda, verilerimizle uyumlu \u00e7al\u0131\u015facak algoritmalar\u0131 se\u00e7meye \u00e7al\u0131\u015ft\u0131k. \u00d6ncelikle, \u00f6neri algoritmalar\u0131n\u0131 besleyecek olan giri\u015f verisini kontrol etmek \u00f6nemlidir. Bu durumda, AKA verileri, dolay\u0131s\u0131yla se\u00e7ilen veri k\u00fcmelerinin verilerini, kullan\u0131c\u0131lar\u0131n \u00f6\u011frenme kaynaklar\u0131 (\u00f6geler) ile etkile\u015fim verilerini i\u00e7erir. Bu nedenle, tavsiye sistemleri ailesinden \u0130\u015fbirlikli Filtreleme'den (\u0130F) faydalanmay\u0131 uygun g\u00f6rd\u00fck. \u0130F algoritmalar\u0131, i\u00e7erik esasl\u0131 tavsiye sistemleri taraf\u0131ndan kullan\u0131lan i\u00e7erik verilerinden daha ziyade kullan\u0131c\u0131lar\u0131n derecelendirmeler, yer imleri, g\u00f6r\u00fcn\u00fcmler, be\u011feniler, vb. gibi etkile\u015fim verilerine dayan\u0131r. \u0130F \u00f6nerileri \u201ct\u00fcr\u00fcne\u201d g\u00f6re bellek tabanl\u0131 veya model tabanl\u0131 olabilir; \u201cteknik\u201de at\u0131fta bulunarak madde baz\u0131nda veya kullan\u0131c\u0131 tabanl\u0131 olabilirler. Bu ayr\u0131mlar\u0131n ayr\u0131nt\u0131l\u0131 bir a\u00e7\u0131klamas\u0131 i\u00e7in B\u00f6l\u00fcm 4 Fazeli vd. (2014) bak\u0131n\u0131z. \u00c7al\u0131\u015fmam\u0131zda, t\u00fcm bellek t\u00fcrlerinden ve tekniklerinden faydaland\u0131k: bellek temelli, model temelli,kullan\u0131c\u0131 temelli ve \u00f6\u011fe temelli. \u015eekil 20.1, \u00fc\u00e7 temel ad\u0131mdan olu\u015fan deneysel y\u00f6ntemimizi g\u00f6stermektedir:<\/span><\/p>\n \n

      \"\"<\/p>\n\u015eekil 20.1. Fazeli vd. (2014)'nin kulland\u0131klar\u0131 deneysel y\u00f6ntem.<\/i><\/span><\/span>\n

        \n \t
      1. \n

        Bellek bazl\u0131 \u0130F'lerin performans\u0131n\u0131 hem kullan\u0131c\u0131 hem de \u00fcr\u00fcn baz\u0131nda olmak \u00fczere farkl\u0131 benzerlik i\u015flevlerini kullanarak kar\u015f\u0131la\u015ft\u0131rd\u0131k.<\/span><\/p>\n<\/li>\n \t

      2. \n

        \u00d6rnek verilerimiz \u00fczerine en geli\u015fmi\u015f Matris \u00c7arpanlar\u0131na Ay\u0131rma y\u00f6ntemleri de d\u00e2hil olmak \u00fczere model tabanl\u0131 \u0130F'leri i\u015fledik.<\/span><\/p>\n<\/li>\n \t

      3. \n

        Ad\u0131m 1 ve 2'den en iyi performans g\u00f6steren algoritmalar\u0131n son bir kar\u015f\u0131la\u015ft\u0131rmas\u0131n\u0131 yapt\u0131k. Dayanaklara ek olarak, en yak\u0131n k kom\u015fusu (kNN) y\u00f6ntemini kullanarak kom\u015fu bulma mekanizmas\u0131n\u0131 geli\u015ftirmek i\u00e7in \u00f6nerilen grafik tabanl\u0131 bir yakla\u015f\u0131m\u0131 de\u011ferlendirdik (Fazeli vd., 2014).<\/span><\/p>\n<\/li>\n<\/ol>\n

        Performans De\u011ferlendirmesi. <\/span>Uygun veri k\u00fcmeleri ve \u00f6neri algoritmalar\u0131n\u0131 se\u00e7tikten sonra, aday algoritmalar\u0131n performans\u0131n\u0131 de\u011ferlendirme g\u00f6revine ula\u015f\u0131r\u0131z. Bunun i\u00e7in bir de\u011ferlendirme protokol\u00fc tan\u0131mlamam\u0131z gerekir (Herlocker vd. 2004). Bir de\u011ferlendirme protokol\u00fcn\u00fcn iyi bir a\u00e7\u0131klamas\u0131 a\u015fa\u011f\u0131daki sorular \u00fczerine e\u011filmelidir:<\/span><\/p>\n\n

        S1. Ne \u00f6l\u00e7\u00fclecek?<\/span><\/span><\/h4>\n

        Genel olarak, \u00e7o\u011fu \u00e7evrimd\u0131\u015f\u0131 tavsiye sistemi \u00e7al\u0131\u015fmalar\u0131nda, \u00fcretilen \u00f6nerilerin kestirim<\/i><\/span> do\u011frulu\u011funu \u00f6l\u00e7\u00fcyoruz. Bununla, bir e\u011fitim setini ve bir test setini kar\u015f\u0131la\u015ft\u0131rarak, derecelendirme kestirimlerinin ger\u00e7ek kestirimlerden ne kadar farkl\u0131 oldu\u011funu \u00f6l\u00e7mek istiyoruz. E\u011fitim ve test setleri, kullan\u0131c\u0131 derecelendirme verilerimizin (kullan\u0131c\u0131 etkile\u015fimi verileriyle ayn\u0131) b\u00f6l\u00fcnmesinden kaynaklanmaktad\u0131r. TD\u00d6AK 2014 \u00e7al\u0131\u015fmam\u0131zda, e\u011fitim seti ve test seti i\u00e7in kullan\u0131c\u0131 derecelendirmelerini s\u0131ras\u0131yla %80 ve %20'ye b\u00f6ld\u00fck. Bu t\u00fcr bir b\u00f6l\u00fcnme tavsiye sistem de\u011ferlendirmelerinde yayg\u0131n olarak kullan\u0131lmaktad\u0131r (Fazeli vd., 2014).<\/span><\/p>\n \n\n\"\"\n\n\u015eekil 20.2. Grafik tabanl\u0131 \u0130F'nin F1'i ve kullan\u0131lan t\u00fcm veri k\u00fcmeleri i\u00e7in en iyi performans g\u00f6steren temel bellek tabanl\u0131 ve model tabanl\u0131 \u0130F'ler (Fazeli vd., 2014).<\/span><\/i><\/span><\/span>\n

        S2. Bir \u00f6neri sistemi \u00e7al\u0131\u015fmas\u0131 i\u00e7in uygun olan \u00f6l\u00e7\u00fcmler nelerdir?<\/span><\/span><\/h4>\n

        Girdi verilerimiz 5 y\u0131ld\u0131zl\u0131 derecelendirmeler gibi a\u00e7\u0131k kullan\u0131c\u0131 tercihleri \u200b\u200b<\/span>i\u00e7eriyorsa, MHO (mutlak hatalar ortalamas\u0131) veya KOKH (k\u00f6k ortalama kare hatas\u0131) kullanabiliriz. MHO ve KOKH'nin her ikisi de kullan\u0131c\u0131 derecelendirmeleriyle ayn\u0131 aral\u0131\u011f\u0131 takip eder; \u00f6rne\u011fin, e\u011fer veriler 5 y\u0131ld\u0131zl\u0131 derecelendirmeler i\u00e7eriyorsa, bu metrikler 1 ile 5 aras\u0131ndad\u0131r.<\/span><\/p>\n

        Girdi verileri g\u00f6r\u00fcn\u00fcmler, yer imleri, indirmeler vb. gibi kesin kullan\u0131c\u0131 tercihlerini i\u00e7eriyorsa Keskinlik, Hassasiyet ve F1 skorlar\u0131n\u0131 kullanabiliriz. F1 skorunu, \u00fcretilen \u00f6nerilerin do\u011frulu\u011funu ve kapsam\u0131n\u0131 de\u011ferlendirmede \u00f6nemli \u00f6l\u00e7\u00fctler olan keskinlik ve hassasiyeti birle\u015ftirdi\u011fi i\u00e7in kulland\u0131k (Herlocker vd., 2004). F1, 0 ile 1 aras\u0131ndad\u0131r.<\/span><\/p>\n

        Ek olarak n' yi metrik olarak \u00f6l\u00e7\u00fclen, kesme<\/i><\/span> olarak da bilinen, en \u00fcstteki n tavsiyelerinde tan\u0131mlamam\u0131z gerekir.<\/span><\/p>\n

        Son olarak tan\u0131mlanan de\u011ferlendirme protokol\u00fcn\u00fc takip ederek veri k\u00fcmelerindeki aday algoritmalar\u0131n \u00e7al\u0131\u015ft\u0131r\u0131lmas\u0131yla ilgili sonu\u00e7lar\u0131 sunuyoruz. S\u0131n\u0131rl\u0131 alan nedeniyle sadece TD\u00d6AK 2014 makalemizin son sonu\u00e7lar\u0131n\u0131 burada sunuyoruz. Daha fazla sonu\u00e7 i\u00e7in l\u00fctfen orijinal makalenin 5.1 ve 5.2 b\u00f6l\u00fcmlerine (Fazeli vd., 2014) bak\u0131n\u0131z.<\/span><\/p>\n

        \u015eekil 20.2, en iyi performans g\u00f6steren bellek tabanl\u0131 \u0130F (Jaccard kNN), \u00e7izge tabanl\u0131 \u0130F ile kar\u015f\u0131la\u015ft\u0131r\u0131ld\u0131\u011f\u0131nda model tabanl\u0131 \u0130F (Bayes\u00e7i y\u00f6ntemi) in F1 sonu\u00e7lar\u0131n\u0131 g\u00f6stermektedir. X ekseni kullan\u0131lan veri k\u00fcmelerini ve y ekseni F1 de\u011ferlerini g\u00f6stermektedir. \u015eekil 20.2'nin g\u00f6sterdi\u011fi gibi \u00e7izge tabanl\u0131 yakla\u015f\u0131m AM\u0130\u00dcV ve MovieLens i\u00e7in en iyi performans\u0131 (%24) g\u00f6sterir ve se\u00e7ilen bellek -tabanl\u0131 ve model- tabanl\u0131 \u0130F'ler, \u00e7izge tabanl\u0131 \u0130F'den hem en sonra ikinci ve \u00fc\u00e7\u00fcnc\u00fc s\u0131rada yer almaktad\u0131r. OpenScout i\u00e7in, bellek tabanl\u0131 yakla\u015f\u0131m neredeyse %1 farkla daha iyi performans g\u00f6sterir.<\/span><\/p>\n

        Sonu\u00e7 olarak \u015fekil 20.2'de sunulan sonu\u00e7lara g\u00f6re \u00e7izge tabanl\u0131 yakla\u015f\u0131m\u0131n sosyal \u00f6\u011frenme platformlar\u0131 i\u00e7in etkili oldu\u011fu g\u00f6r\u00fclmektedir. Bu keskinlik ve yap\u0131lan \u00f6nerinin hassasiyetini etkili bir birle\u015fimi olan geli\u015fmi\u015f bir F1 ile yans\u0131t\u0131l\u0131r.<\/span><\/p>\n\n

        \u00d6neri Sisteminin Uygulanmas\u0131 ve Kullan\u0131c\u0131 \u00c7al\u0131\u015fmas\u0131<\/h3>\n

        E\u011fitim alan\u0131nda, kullan\u0131c\u0131 \u00e7al\u0131\u015fmalar\u0131n\u0131n \u00f6nemi her zamankinden daha da belirgin hale gelmi\u015ftir (Drachsler vd., 2015). E\u011fitimde tavsiye sistemlerinin temel amac\u0131 do\u011fru kestirimlerin \u00e7ok daha fazlas\u0131n\u0131 i\u00e7erdi\u011fi i\u00e7in fayda, yenilik ve \u00f6nerilerin \u00e7e\u015fitlili\u011fi gibi di\u011fer kalite g\u00f6stergelerini de yap\u0131land\u0131rmal\u0131d\u0131r. Bununla birlikte, tavsiye sistemi \u00e7al\u0131\u015fmalar\u0131n\u0131n \u00e7o\u011funlu\u011fu hala yaln\u0131zca \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmalar\u0131na dayanmaktad\u0131r. Bunun nedeni muhtemelen kullan\u0131c\u0131 \u00e7al\u0131\u015fmalar\u0131n\u0131n zaman al\u0131c\u0131 ve karma\u015f\u0131k olmas\u0131d\u0131r.<\/span><\/p>\n

        AKA verileri \u00fczerinde \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmas\u0131 yapt\u0131ktan sonra hedef platformumuzda bir kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131 y\u00fcr\u00fcten Fazeli vd. (2014) a\u00e7\u0131klad\u0131\u011f\u0131 \u00e7al\u0131\u015fmalar\u0131 daha da ileriye ta\u015f\u0131d\u0131k. Bunun i\u00e7in en iyi performans g\u00f6steren algoritmalar\u0131 AKA ile birle\u015ftirdik. Ger\u00e7ek AKA kullan\u0131c\u0131lar\u0131na, onlar i\u00e7in yapt\u0131\u011f\u0131m\u0131z \u00f6nerilerden memnun olup olmad\u0131klar\u0131n\u0131 sorduk. Bunun i\u00e7in \u015fu be\u015f fakt\u00f6r\u00fc kullanarak k\u0131sa bir anket haz\u0131rlad\u0131k: kullan\u0131\u015fl\u0131l\u0131k, do\u011fruluk, yenilik, \u00e7e\u015fitlilik ve tesad\u00fcfen de\u011ferli bir \u015feyler ke\u015ffetme yetene\u011fi. Bu veri \u00e7al\u0131\u015fmas\u0131n\u0131n ve takip eden kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131n\u0131n tam a\u00e7\u0131klamas\u0131 ve sonu\u00e7lar\u0131 hen\u00fcz yay\u0131nlanmam\u0131\u015ft\u0131r. Kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131, ger\u00e7ek AKA verileri \u00fczerinde y\u00fcr\u00fctt\u00fc\u011f\u00fcm\u00fcz veri \u00e7al\u0131\u015fmas\u0131n\u0131n sonu\u00e7lar\u0131n\u0131 do\u011frulamaz, tahmin do\u011frulu\u011fu gibi veri \u00e7al\u0131\u015fmalar\u0131n\u0131n ba\u015far\u0131 g\u00f6stergelerinin \u00f6tesine ge\u00e7ebilecek kullan\u0131c\u0131 \u00e7al\u0131\u015fmalar\u0131 yapman\u0131n olduk\u00e7a gerekli oldu\u011funu g\u00f6sterir.<\/span><\/p>\n

        Do\u011fruluk, tavsiye sistemlerini de\u011ferlendirmedeki \u00f6nemli \u00f6l\u00e7\u00fctlerden biridir ancak sadece bu metri\u011fe g\u00fcvenmek veri bilimcilerinin ve e\u011fitim teknologlar\u0131n\u0131n daha az etkili yollara y\u00f6nelmelerine neden olabilir.<\/span><\/p>\n\n

        PRAT\u0130K \u00c7IKARIMLAR VE SINIRLAMALAR<\/h2>\n

        E\u011fitsel veri k\u00fcmeleri, referans ba\u011flant\u0131lar\u0131 arac\u0131l\u0131\u011f\u0131yla halka a\u00e7\u0131k ve a\u00e7\u0131k\u00e7a kullan\u0131labilir olmad\u0131klar\u0131ndan e\u011fitim veri k\u00fcmelerinin \u00e7o\u011funa eri\u015fmek zordur. Dahas\u0131 ilgili \u00e7al\u0131\u015fmalara ait bulgular\u0131n kar\u015f\u0131la\u015ft\u0131r\u0131lmas\u0131 zordur, \u00f6rne\u011fin Verbert vd. (2011) ve Manouselis, Vuorikari, and Van Assche\u2019 nin (2010) \u00e7al\u0131\u015fmalar\u0131 gibi. Her ne kadar ayn\u0131 veri k\u00fcmelerini ve bu iki \u00e7al\u0131\u015fmada kullan\u0131lan baz\u0131 algoritmalar\u0131 uygulasak da \u00f6rnek \u00e7al\u0131\u015fmam\u0131z\u0131n sonu\u00e7lar\u0131 onlar\u0131n sonu\u00e7lar\u0131ndan farkl\u0131d\u0131r. Bu nedenle, \u00f6\u011frenme kaynaklar\u0131n\u0131n ki\u015fiselle\u015ftirilmesine ili\u015fkin kar\u015f\u0131la\u015ft\u0131rmalardan ek bilgi elde edemedik. Olas\u0131 sebeplerden biri \u00e7al\u0131\u015fmalar\u0131n ayn\u0131 veri k\u00fcmesinin farkl\u0131 versiyonlar\u0131n\u0131 kullanmas\u0131d\u0131r \u00e7\u00fcnk\u00fc toplanan veriler farkl\u0131 zamanlara aittir. \u00d6rne\u011fin AM\u0130\u00dcV veri k\u00fcmesi i\u00e7in farkl\u0131 s\u00fcr\u00fcmler mevcuttur. Asl\u0131nda, denemeler yapmak i\u00e7in ya da tavsiye sistemi toplulu\u011funda kar\u015f\u0131la\u015ft\u0131rma yapmak i\u00e7in benzersiz bir s\u00fcr\u00fcm belirlenmemi\u015ftir.<\/span><\/p>\n

        Bu sorun, maalesef e\u011fitim alan\u0131nda e-ticaret d\u00fcnyas\u0131nda bulunan MovieLens veri k\u00fcmesi1<\/a><\/sup> ile kar\u015f\u0131la\u015ft\u0131r\u0131labilecek alt\u0131n bir standart veri k\u00fcmesinin olmamas\u0131ndan kaynaklanmaktad\u0131r. Asl\u0131nda, \u00d6A toplulu\u011fu, farkl\u0131 ki\u015fiselle\u015ftirme yakla\u015f\u0131mlar\u0131 i\u00e7in ana referanslar seti olarak kullan\u0131labilecek birka\u00e7 temsili veri k\u00fcmesine ihtiya\u00e7 duymaktad\u0131r. Temel ama\u00e7, \u00d6A ara\u015ft\u0131rmas\u0131n\u0131 y\u00fcr\u00fctmek i\u00e7in standart bir veri format\u0131 elde etmektir. Bu fikir ba\u015flang\u0131\u00e7ta dataTEL projesi (Drachsler vd., 2011) taraf\u0131ndan \u00f6nerilmi\u015f ve daha sonra SoLAR \u00d6\u011frenme Analiti\u011fi Toplulu\u011fu taraf\u0131ndan takip edilmi\u015ftir (Ga\u0161evi\u0107 vd., 2011). KA\u00c7D'ler alan\u0131nda, Drachsler ve Kalz (2016), bu kar\u015f\u0131la\u015ft\u0131r\u0131labilir sonu\u00e7 eksikli\u011fini ve bilimsel sonu\u00e7lar\u0131 kar\u015f\u0131la\u015ft\u0131rmak i\u00e7in veri havuzlar\u0131n\u0131 kullanan bir ara\u015ft\u0131rma d\u00f6ng\u00fcs\u00fcne duyulan acil ihtiyac\u0131 tart\u0131\u015fm\u0131\u015flard\u0131r. Dahas\u0131, AB taraf\u0131ndan finanse edilen LinkedUp2<\/a><\/sup> adl\u0131 bir proje, ba\u011fl\u0131 veri kavramlar\u0131n\u0131 uygulayarak bir dizi alt\u0131n standart veri k\u00fcmesini sa\u011flama y\u00f6n\u00fcnde \u00fcmit verici bir yakla\u015f\u0131m izlemektedir (Bizer, Heath ve Berners \u2013 Lee, 2009). LinkedUp projesi, \u00f6\u011frenme analiti\u011fi ara\u015ft\u0131rmalar\u0131 i\u00e7in ba\u011flant\u0131l\u0131 bir veri havuzu sa\u011flamay\u0131 ve merkezi veri havuzu \u00fczerinden \u00e7e\u015fitli veri yar\u0131\u015fmalar\u0131n\u0131 y\u00fcr\u00fctmeyi ama\u00e7l\u0131yor.<\/span><\/p>\n

        Genel olarak, farkl\u0131 tavsiye sistemlerinin sonu\u00e7lar\u0131 veya e\u011fitim alan\u0131ndaki ki\u015fiselle\u015ftirme yakla\u015f\u0131mlar\u0131n\u0131n kar\u015f\u0131la\u015ft\u0131r\u0131lmas\u0131 algoritmalar\u0131n, \u00f6\u011frenen modellerinin, veri k\u00fcmelerinin ve de\u011ferlendirme \u00f6l\u00e7\u00fctlerinin \u00e7e\u015fitlili\u011fi nedeniyle hala g\u00fc\u00e7l\u00fckle yap\u0131labilmektedir(Drachsler vd., 2015; Manouselis vd., 2012).<\/span><\/p>\n\n

        SONU\u00c7<\/h2>\n

        Bu b\u00f6l\u00fcm\u00fcn temel amac\u0131 bir \u00f6\u011frenme ortam\u0131 i\u00e7in en uygun tavsiye sisteminin nas\u0131l tan\u0131mland\u0131\u011f\u0131n\u0131 g\u00f6stermektir. Bunu yapmak i\u00e7in, Drachsler vd.nin (2015) \u00f6\u011frenmede dan\u0131\u015fman sistemlerini de\u011ferlendirmek i\u00e7in sunduklar\u0131 standart metodolojiyi kullanarak bir \u00f6rnek veri \u00e7al\u0131\u015fmas\u0131n\u0131 takip ettik. Metodoloji d\u00f6rt temel basamaktan olu\u015fur:<\/span><\/p>\n\n

          \n \t
        1. \n

          Ger\u00e7ek verinin hen\u00fcz mevcut olmamas\u0131 durumu g\u00f6z \u00f6n\u00fcnde bulundurularak hedef verilerin benzeri olan ve tercihen de e\u011fitim alan\u0131ndan uygun veri k\u00fcmelerini se\u00e7in.<\/span><\/p>\n<\/li>\n \t

        2. \n

          Girdi verilerine en uygun olan bir dizi aday \u00f6neri algoritmalar\u0131n\u0131 \u00e7al\u0131\u015ft\u0131r\u0131n. Bu basama\u011f\u0131n \u00e7\u0131kt\u0131s\u0131 girdi verileriyle en iyi \u015fekilde \u00e7al\u0131\u015facak \u00f6neri algoritmalar\u0131n\u0131 ortaya \u00e7\u0131karmal\u0131d\u0131r.<\/span><\/p>\n<\/li>\n \t

        3. \n

          Kendileri i\u00e7in yap\u0131lan tavsiyelerde kullan\u0131c\u0131 memnuniyetini \u00f6l\u00e7mek amac\u0131yla bir kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131 ger\u00e7ekle\u015ftirin.<\/span><\/p>\n<\/li>\n \t

        4. \n

          En iyi aday tavsiyeyi hedef \u00f6\u011frenme platformuna da\u011f\u0131t\u0131n.<\/span><\/p>\n<\/li>\n<\/ol>\n

          Kullan\u0131c\u0131lara y\u00f6nelik yapt\u0131\u011f\u0131m\u0131z \u00e7al\u0131\u015fma sonu\u00e7lar\u0131n\u0131n, \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmas\u0131 sonu\u00e7lar\u0131n\u0131 do\u011frulamam\u0131\u015f olmas\u0131 ger\u00e7e\u011fi, her ne kadar zaman al\u0131c\u0131 ve karma\u015f\u0131k olsa da kullan\u0131c\u0131lara y\u00f6nelik \u00e7al\u0131\u015fmalar y\u00fcr\u00fct\u00fclmesinin \u00f6nemini ortaya koymaktad\u0131r.<\/span><\/p>\n\n

          TE\u015eEKK\u00dcR B\u00d6L\u00dcM\u00dc<\/h2>\n

          Bu b\u00f6l\u00fcm k\u0131smen AB 7. \u00c7er\u00e7eve Program\u0131 A\u00e7\u0131k Uzay Ara\u015ft\u0131rmalar\u0131 projesi kapsam\u0131nda finanse edilmi\u015ftir. Bu belge Avrupa Birli\u011fi'nin g\u00f6r\u00fc\u015flerini temsil etmemektedir ve Avrupa Birli\u011fi i\u00e7eri\u011finden kaynaklanabilecek herhangi bir kullan\u0131mdan sorumlu de\u011fildir. Hendrik Drachsler'in \u00e7al\u0131\u015fmalar\u0131 FP7 AB projesi \u00d6ATP taraf\u0131ndan desteklenmi\u015ftir.<\/span><\/p>\n\n

          KAYNAK\u00c7A<\/h2>\nBizer, C., Heath, T., & Berners-Lee, T. (2009). Linked data: The story so far. International Journal on Semantic Web and Information Systems, 5<\/i>(3), 1\u201322. <\/span><\/span>\n\nDietze, S., Siemens, G., Taibi, D., & Drachsler, H. (2016). Editorial: Datasets for learning analytics. Journal of Learning Analytics, 3<\/i>(3), 307\u2013311. http:\/\/dx.doi.org\/10.18608\/jla.2016.32.15<\/span><\/span>\n\nDrachsler, H., Hummel, H., van den Berg, B., Eshuis, J., Waterink, W., Nadolski, R., Berlanga, A., Boers, N., & Koper, R. (2009). Effects of the ISIS recommender system for navigation support in self-organised learning networks. Journal of Educational Technology & Society, 12<\/i>(3), 115\u2013126. <\/span><\/span>\n\nDrachsler, H., & Kalz, M. (2016). The MOOC and learning analytics innovation cycle (MOLAC): A reflective summary of ongoing research and its challenges. Journal of Computer Assisted Learning, 32<\/i>(3), 281\u2013290. http:\/\/ doi.org\/ 10.1111\/jcal.12135 <\/span><\/span>\n\nDrachsler, H., Rutledge, L., Van Rosmalen, P., Hummel, H., Pecceu, D., Arts, T., Hutten, E., & Koper, R. (2010). ReMashed: An usability study of a recommender system for mash-ups for learning. International Journal of Emerging Technologies in Learning, 5<\/i>. http:\/\/online-journals.org\/i-jet\/article\/view\/1191 <\/span><\/span>\n\nDrachsler, H., Verbert, K., Santos, O., & Manouselis, N. (2015). Panorama of recommender systems to support learning. In F. Ricci, L. Rokach, & B. Shapira (Eds.), Recommender systems handbook<\/i>, 2nd ed. Springer. <\/span><\/span>\n\nDrachsler, H., Verbert, K., Sicilia, M.-A., Wolpers, M., Manouselis, N., Vuorikari, R., Lindstaedt, S., & Fischer, F. (2011). dataTEL: Datasets for Technology Enhanced Learning \u2014 White Paper<\/i>. Stellar Open Archive. http: \/\/ dspace.ou.nl\/handle\/1820\/3846 <\/span><\/span>\n\nDuval, E. (2011). Attention please! Learning analytics for visualization and recommendation. Proceedings of the 1st International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201911), 27 February\u20131 March 2011, Banff, AB, Canada (pp. 9\u201317). New York: ACM. <\/span><\/span>\n\nFazeli, S., Loni, B., Drachsler, H., & Sloep, P. (2014). Which recommender system can best fit social learning platforms? Proceedings of the 9th European Conference on Technology Enhanced Learning <\/i>(EC-TEL\u201914), 16\u201319 September 2014, Graz, Austria (pp. 84\u201397). <\/span><\/span>\n\nGa\u0161evi\u0107, G., Dawson, C., Ferguson, S. B., Duval, E., Verbert, K., & Baker, R. S. J. d. (2011, July 28). Open learning analytics: An integrated and modularized platform<\/i>. http:\/\/www.elearnspace.org\/blog\/wp-content\/uploads\/2016\/02\/ProposalLearningAnalyticsModel_SoLAR.pdf <\/span><\/span>\n\nGreller, W., & Drachsler, H. (2012). Translating learning into numbers: A generic framework for learning analytics. Proceedings of the 2nd International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201912), 29 April\u20132 May 2012, Vancouver, BC, Canada (pp. 42\u201357). New York: ACM. <\/span><\/span>\n\nHerlocker, J. L., Konstan, J. A., Terveen, L. G., & Riedl, J. T. (2004). Evaluating collaborative filtering recommender systems. ACM Transactions on Information Systems, 22<\/i>(1), 5\u201353. <\/span><\/span>\n\nManouselis, N., Drachsler, H., Verbert, K., & Duval, E. (2012). Recommender systems for learning<\/i>. Springer Berlin Heidelberg. <\/span><\/span>\n\nManouselis, N., Vuorikari, R., & Van Assche, F. (2010). Collaborative recommendation of e-learning resources: An experimental investigation. Journal of Computer Assisted Learning, 26<\/i>(4), 227\u2013242. <\/span><\/span>\n\nPazzani, M. J., & Billsus, D. (2007). Content-based recommendation systems. In P. Brusilovsky, A. Kobsa, & W. Nejdl (Eds.), The adaptive web: Methods and strategies of web personalization <\/i>(pp. 325\u2013341). Lecture Notes in Computer Science vol. 4321. Springer. <\/span><\/span>\n\nSchafer, J. B., Frankowski, D., Herlocker, J., & Sen, S. (2007). Collaborative filtering recommender systems. In P. Brusilovsky, A. Kobsa, & W. Nejdl (Eds.), The adaptive web: Methods and strategies of web personalization <\/i>(pp. 291\u2013324). Lecture Notes in Computer Science vol. 4321. Springer. <\/span><\/span>\n\nSchmitz, H., Scheffel, M., Friedrich, M., Jahn, M., Niemann, K., Wolpers, M., & Augustin, S. (2009). CAMera for PLE. In U. Cress, V. Dimitrova, & M. Specht (Eds.), Learning in the Synergy of Multiple Disciplines: 4th European Conference on Technology Enhanced Learning (EC-TEL 2009) Nice, France, September 29\u2013October 2, 2009 Proceedings <\/i>(pp. 507\u2013520). Lecture Notes in Computer Science vol. 5794. Springer. <\/span><\/span>\n\nVerbert, K., Drachsler, H., Manouselis, N., Wolpers, M., Vuorikari, R., & Duval, E. (2011). Dataset-driven research for improving recommender systems for learning. Proceedings of the 1st International Conference on Learning Analytics and Knowledge (LAK\u201911), 27 February\u20131 March 2011, Banff, AB, Canada (pp. 44\u201353). New York: ACM.<\/span><\/span>\n\n \n\n
          \n\n
          \n

          1<\/a> https:\/\/epress.lib.uts.edu.au\/journals\/index.php\/JLA\/article\/view\/5071\/5600<\/span><\/p>\n\n<\/div>\n

          \n

          2<\/a> http:\/\/opendiscoveryspace.eu<\/span><\/p>\n\n<\/div>\n

          \n

          3<\/a> http:\/\/www.grouplens.org\/node\/73<\/span><\/p>\n\n<\/div>\n

          \n

          4<\/a> www.linkedup-project.eu<\/span><\/p>\n\n<\/div>\n","rendered":"

          Soude Fazeli1<\/sup>, Hendrik Drachsler2<\/sup>, Peter Sloep2<\/sup><\/span><\/span><\/p>\n

          1<\/sup>Welten Enstit\u00fcs\u00fc, \u00d6\u011frenme, \u00d6\u011fretme ve Teknoloji Ara\u015ft\u0131rma Merkezi, Hollanda<\/span><\/span><\/p>\n

          2<\/sup>Hollanda A\u00e7\u0131k \u00dcniversitesi, Hollanda<\/span><\/span><\/p>\n

          DOI: 10.18608\/hla17.020<\/span><\/span><\/p>\n

          \u00d6Z<\/h2>\n

          Bu b\u00f6l\u00fcm, bir \u00f6neri sistem deneyiminin \u00f6\u011frenme analiti\u011fi (\u00d6A) alan\u0131nda nas\u0131l uygulanabilece\u011fine ili\u015fkin bir \u00f6rnek sunmaktad\u0131r. Bu b\u00f6l\u00fcmde sunulan \u00f6rnek \u00e7al\u0131\u015fma, \u00f6\u011frenmede \u00f6neri sistemlerini de\u011ferlendirmek i\u00e7in standart bir y\u00f6ntem izlemi\u015ftir. \u00d6rnek, Avrupa’daki e\u011fitim payda\u015flar\u0131 i\u00e7in, Facebook benzeri bir sosyal \u00f6\u011frenme platformu sa\u011flamay\u0131 ancak Facebook’ dan farkl\u0131 olarak sadece \u00f6\u011frenme ve bilgi payla\u015f\u0131m\u0131n\u0131 ama\u00e7layan, FP7 programlar\u0131nda A\u00e7\u0131k Ke\u015fif Alan\u0131 (AKA) projesi kapsam\u0131nda haz\u0131rlanm\u0131\u015ft\u0131r. Bu b\u00f6l\u00fcmde, ad\u0131m ad\u0131m bir s\u00fcre\u00e7te tam bir tavsiye sistemi veri \u00e7al\u0131\u015fmas\u0131n\u0131 a\u00e7\u0131klamaktay\u0131z. Ayr\u0131ca, \u00f6\u011frenme alan\u0131ndaki veri g\u00fcd\u00fcml\u00fc \u00e7al\u0131\u015fmalar\u0131n eksikliklerini ana hatlar\u0131yla belirtmekte ve SoLAR toplulu\u011fu taraf\u0131ndan \u00f6nerildi\u011fi gibi a\u00e7\u0131k bir \u00f6\u011frenme analiti\u011fi platformuna olan ihtiyac\u0131 vurgulamaktay\u0131z.<\/span><\/span><\/p>\n

          Anahtar Kelimeler<\/span>: Tavsiye sistemi, \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmas\u0131, kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131, i\u015fbirlikli filtreleme, bilgi arama ve alma, bilgi filtreleme, ara\u015ft\u0131rma, metodoloji, seyreklik <\/span><\/span><\/p>\n

          \u00c7e\u015fitli alanlarda b\u00fcy\u00fck miktarlarda verinin ortaya \u00e7\u0131kmas\u0131yla birlikte, tavsiye sistemleri, kullan\u0131c\u0131lar\u0131n ge\u00e7mi\u015f davran\u0131\u015flar\u0131na ve mevcut durumlar\u0131na g\u00f6re en uygun bilgileri sa\u011flamak i\u00e7in pratik bir yakla\u015f\u0131m haline gelmi\u015ftir. Duval (2011) tavsiye sistemlerini “‘Se\u00e7im paradoksunu’ ele almak ve bollu\u011fu bir problemden \u00f6\u011frenme i\u00e7in bir varl\u0131\u011fa d\u00f6n\u00fc\u015ft\u00fcrmek\u201din bir \u00e7\u00f6z\u00fcm\u00fc olarak ortaya koymu\u015ftur (s. 9), e\u011fitsel veri madencili\u011fi, b\u00fcy\u00fck veri ve web analiti\u011fi gibi alanlar\u0131n t\u00fcm\u00fc b\u00fcy\u00fck miktarda veride \u00f6r\u00fcnt\u00fc bulmaya \u00e7al\u0131\u015ft\u0131\u011f\u0131na dikkat \u00e7ekiyor. \u00d6rne\u011fin, veri madencili\u011fi yakla\u015f\u0131mlar\u0131, kullan\u0131c\u0131lar\u0131n toplanan verilerinden tespit edilen benzerlik \u00f6r\u00fcnt\u00fclerine dayanarak tavsiyelerde bulunabilir. Bunun yan\u0131 s\u0131ra, Greller ve Drachsler (2012) taraf\u0131ndan yap\u0131lan bir saha ara\u015ft\u0131rmas\u0131, tavsiye sistemlerini ve ki\u015fiselle\u015ftirmeyi \u00d6A ara\u015ft\u0131rmas\u0131n\u0131n \u00f6nemli bir par\u00e7as\u0131 olarak tan\u0131mlam\u0131\u015ft\u0131r.<\/span><\/p>\n

          Tavsiye sistemleri teknolojilerinin temeline ve algoritmalar\u0131na g\u00f6re ay\u0131rt edilebilir. Kabaca i\u00e7erik tabanl\u0131 ya da i\u015fbirlikli filtreleme kullan\u0131rlar. \u0130\u00e7erik tabanl\u0131 algoritmalar, tavsiye sistemlerinde kullan\u0131lan ana y\u00f6ntemlerden biridir; \u00f6genin i\u00e7eri\u011finin temsilini kullan\u0131c\u0131n\u0131n tercih modeliyle kar\u015f\u0131la\u015ft\u0131rarak kullan\u0131c\u0131ya bir \u00f6ge tavsiye ederler (Pazzani ve Billsus, 2007). Ortak filtreleme, kullan\u0131c\u0131lar\u0131n \u00f6geleri hakk\u0131ndaki g\u00f6r\u00fc\u015flerine ve geri bildirimlerine dayan\u0131r. \u0130\u015fbirlikli filtreleme algoritmalar\u0131 ilk \u00f6nce benzer d\u00fc\u015f\u00fcnen kullan\u0131c\u0131lar\u0131 bulur ve bunlar\u0131 baz\u0131 hedef kullan\u0131c\u0131lara en yak\u0131n kom\u015fular olarak tan\u0131t\u0131r; daha sonra, bu kullan\u0131c\u0131n\u0131n hedeflenen kullan\u0131c\u0131lar\u0131n en yak\u0131n kom\u015fular\u0131 (e\u015f dereceleri) taraf\u0131ndan verilen \u00f6gelere g\u00f6re bir \u00f6genin derecelendirmesini tahmin ederler (Herlocker, Konstan, Terveen ve Riedl, 2004; Manouselis, Drachsler, Verbert ve Duval, 2012; Schafer, Frankowski, Herlocker ve Sen, 2007).<\/span><\/p>\n

          Ge\u00e7mi\u015fte, farkl\u0131 ama\u00e7larla \u00e7e\u015fitli e\u011fitim projelerinde tavsiye sistemlerini uygulad\u0131k (Drachsler vd., 2010; Fazeli, Loni, Drachsler ve Sloep, 2014; Drachsler vd., 2009). Bu b\u00f6l\u00fcmde, e\u011fitimde tavsiye sistemi algoritmalar\u0131n\u0131n geli\u015ftirilmesi ve de\u011ferlendirilmesi ile ilgili \u015fu ana kadar tespit etti\u011fimiz en iyi uygulamalardan baz\u0131lar\u0131n\u0131 payla\u015fmak istiyoruz; \u00f6zellikle de bir tavsiye sistemleri denemesinin nas\u0131l kurulaca\u011f\u0131 ve \u00e7al\u0131\u015ft\u0131r\u0131laca\u011f\u0131na dair bir \u00f6rnek vermek istiyoruz.<\/span><\/p>\n

          Teknoloji ile Geli\u015ftirilmi\u015f \u00d6\u011frenme alan\u0131na istinaden TavSisTD\u00d6 \u00e7al\u0131\u015fma grubunun Tavsiye Sistemleri i\u00e7in yapt\u0131\u011f\u0131 a\u00e7\u0131klamaya g\u00f6re (Drachsler, Verbert, Santos ve Manouselis) standart bir de\u011ferlendirme y\u00f6nteminin uygulanmas\u0131 \u00f6nemlidir. \u00c7al\u0131\u015fma grubu, e\u011fitimde bir tavsiye sistemini de\u011ferlendirmek i\u00e7in d\u00f6rt kritik ad\u0131mdan olu\u015fan bir ara\u015ft\u0131rma metodolojisi tan\u0131mlam\u0131\u015ft\u0131r:<\/span><\/p>\n

            \n
          1. \n

            Tavsiye g\u00f6revine uyan bir dizi<\/i><\/span> veri k\u00fcmesinin se\u00e7imi. \u00d6rne\u011fin, tavsiye g\u00f6revi bir kullan\u0131c\u0131 i\u00e7in yeni \u00f6geler bulmak veya ilgili \u00f6geleri ke\u015ffetmek olabilir.<\/span><\/p>\n<\/li>\n

          2. \n

            Tavsiye edilen sistemlerin performans\u0131 hakk\u0131nda bilgi vermek i\u00e7in iyi bilinen veri k\u00fcmeleri (m\u00fcmk\u00fcnse<\/i><\/span>, MovieLens gibi e\u011fitime dayal\u0131 veri k\u00fcmeleri d\u00e2hil) d\u00e2hil olmak \u00fczere se\u00e7ilen veri k\u00fcmelerinde farkl\u0131 algoritmalar\u0131n \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmas\u0131.<\/span><\/p>\n<\/li>\n

          3. \n

            Tasarlanan tavsiye sisteminin teknik y\u00f6nlerinin yan\u0131 s\u0131ra \u00f6\u011frenenler \u00fczerindeki psiko-e\u011fitsel etkilerini test etmek i\u00e7in kapsaml\u0131 bir kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131.<\/span><\/p>\n<\/li>\n

          4. \n

            Tavsiye sisteminin ger\u00e7ek kullan\u0131c\u0131lar ile ger\u00e7ek\u00e7i, normal \u00e7al\u0131\u015fma ko\u015fullar\u0131 alt\u0131nda test edilebilen ger\u00e7ek zamanl\u0131 bir uygulamada kullan\u0131lmas\u0131.<\/span><\/p>\n<\/li>\n<\/ol>\n

            Tavsiye sisteminin tam bir a\u00e7\u0131klamas\u0131 yukar\u0131daki d\u00f6rt ad\u0131ma sunulan s\u0131n\u0131fland\u0131rma \u00e7er\u00e7evesine g\u00f6re eklenmelidir. (Drachsler vd., 2015). Kullan\u0131lan veri k\u00fcmesi, \u00d6\u011frenme Analiti\u011fi Dergisi1<\/a><\/sup>‘nin e\u011fitsel veri k\u00fcmeleri ile ilgili \u00f6zel b\u00f6l\u00fcm\u00fcnde bildirilmelidir ve belirli ko\u015fullar alt\u0131nda di\u011fer ara\u015ft\u0131rmac\u0131lar i\u00e7in haz\u0131r bulundurulmal\u0131d\u0131r (Dietze, Siemens, Taibi ve Drachsler, 2016). Bu di\u011fer ara\u015ft\u0131rmac\u0131lar\u0131n kar\u015f\u0131la\u015ft\u0131r\u0131labilir sonu\u00e7lar\u0131 ve yeni g\u00f6r\u00fc\u015fleri elde etmek i\u00e7in ara\u015ft\u0131rman\u0131n herhangi bir b\u00f6l\u00fcm\u00fcn\u00fc tekrarlamas\u0131n\u0131 ve d\u00fczenlemesini ve b\u00f6ylelikle analitik \u00f6\u011frenmenin tavsiye sistemleri gibi bir bilgi birikiminin olu\u015fmas\u0131n\u0131 m\u00fcmk\u00fcn k\u0131lar.<\/span><\/p>\n

            Bu b\u00f6l\u00fcmde e\u011fitimde tavsiye sistemleri i\u00e7in yukar\u0131da a\u00e7\u0131klanan ara\u015ft\u0131rma metodolojisini takip eden deneysel bir \u00e7al\u0131\u015fma \u00f6rne\u011fi sunuyoruz. B\u00f6l\u00fcm\u00fcn geri kalan\u0131 \u015fu \u015fekilde d\u00fczenlenmi\u015ftir: Bu b\u00f6l\u00fcmde, e\u011fitimde tavsiye veren sistemler i\u00e7in yukar\u0131da a\u00e7\u0131klanan ara\u015ft\u0131rma metodolojisini izleyen deneysel bir \u00e7al\u0131\u015fma \u00f6rne\u011fi sunaca\u011f\u0131z. Daha sonra, deneyin pratik sonu\u00e7lar\u0131n\u0131 a\u00e7\u0131klayacak ve sonu\u00e7land\u0131raca\u011f\u0131z.<\/span><\/p>\n

            E\u011e\u0130T\u0130M ALANINDA B\u0130R \u00d6NER\u0130 S\u0130STEM\u0130 DENEY\u0130M\u0130<\/h2>\n

            Bu b\u00f6l\u00fcmde, 2014 TD\u00d6AK \u00e7al\u0131\u015fmam\u0131zda sunulan deneysel bir \u00e7al\u0131\u015fmay\u0131 kullanarak \u00f6\u011frenmede bir tavsiye sisteminin nas\u0131l de\u011ferlendirilmesi gerekti\u011fi a\u00e7\u0131klanmaktad\u0131r (Fazeli vd., 2014). Bu \u00e7al\u0131\u015fma, yukar\u0131da a\u00e7\u0131klanan standart metodolojiyi izler. Bununla birlikte, bu metodolojiye ek bir ad\u0131m daha ekledik: TavSisTD\u00d6’n\u00fcn \u00f6zel bir say\u0131s\u0131nda sunulan (Manouselis vd., 2012), kavramsal \/ teorik bir model geli\u015ftirme (Fazeli vd., 2013).<\/span><\/p>\n

            \u00c7al\u0131\u015fmam\u0131zda hedef \u00e7evremiz genel olarak sosyal \u00f6\u011frenme platformlar\u0131d\u0131r. Sosyal \u00f6\u011frenme platformlar\u0131, Facebook gibi sosyal a\u011flara benzer \u015fekilde \u00e7al\u0131\u015f\u0131r ancak Facebook’un aksine yaln\u0131zca \u00f6\u011frenme ve bilgi payla\u015f\u0131m\u0131 amac\u0131yla geli\u015ftirilmi\u015ftir. Bu nedenle; genellikle \u00f6\u011fretmenler, \u00f6\u011frenciler, \u00f6\u011frenenler, politika yap\u0131c\u0131lar gibi e\u011fitim payda\u015flar\u0131 i\u00e7in ortak bir alan olarak hizmet sunarlar. Hedef sosyal \u00f6\u011frenme platformumuz A\u00e7\u0131k Ke\u015fif Alan\u0131 (AKA)’d\u0131r.2<\/a><\/sup> AKA ana sayfas\u0131nda belirtildi\u011fi gibi,<\/span><\/p>\n

            AKA, e-\u00f6\u011frenme ortam\u0131n\u0131n Avrupa ba\u011flam\u0131nda kar\u015f\u0131la\u015ft\u0131\u011f\u0131 \u00e7e\u015fitli zorluklar\u0131 ele almaktad\u0131r. Aray\u00fcz \u00f6\u011frenciler, \u00f6\u011fretmenler, veliler ve politika yap\u0131c\u0131lar g\u00f6z\u00f6n\u00fcnde bulundurularak tasarlanm\u0131\u015ft\u0131r. AKA \u00fc\u00e7 ana hedefi yerine getirecektir. \u00d6ncelikle, payda\u015flar\u0131, da\u011f\u0131n\u0131k e\u011fitsel veri ambarlar\u0131ndan gelen e-\u00f6\u011frenme kaynaklar\u0131 i\u00e7in tek, entegre bir eri\u015fim noktas\u0131 arac\u0131l\u0131\u011f\u0131yla g\u00fc\u00e7lendirecektir. \u0130kincisi, sosyal a\u011f tarz\u0131 \u00e7ok dilli bir portal kullanarak, e-\u00f6\u011frenme kaynaklar\u0131n\u0131 ve ayn\u0131 zamanda e\u011fitim faaliyetlerinin \u00fcretimi i\u00e7in hizmetler sunarak, anlaml\u0131 e\u011fitim faaliyetlerinin \u00fcretiminde payda\u015flar\u0131 hedef almaktad\u0131r. \u00dc\u00e7\u00fcnc\u00fcs\u00fc, payda\u015flar\u0131n okul e\u011fitiminde benimsemeleri i\u00e7in prototip g\u00f6revi g\u00f6rebilecek yeni e\u011fitim faaliyetlerinin etkisini de\u011ferlendirecektir.<\/span><\/p>\n

            \u00c7al\u0131\u015fmam\u0131z\u0131n temel amac\u0131, hangi tavsiye sisteminin bir sosyal \u00f6\u011frenme platformunun veri ve bilgi ihtiya\u00e7lar\u0131na en iyi \u015fekilde cevap verebilece\u011fini bulmakt\u0131r, as\u0131l tavsiye g\u00f6revi kullan\u0131c\u0131lar i\u00e7in ilgili \u00f6geleri bulmakt\u0131r. A\u015fa\u011f\u0131daki alt b\u00f6l\u00fcmlerde, \u00e7al\u0131\u015fmayla ilgili detaylar ad\u0131m ad\u0131m verilmektedir.<\/span><\/p>\n

            Veri K\u00fcmesi Se\u00e7imi<\/h3>\n

            \u00c7o\u011fu veri \u00e7al\u0131\u015fmas\u0131 belirli bir ortam\u0131 veya belirli bir kullan\u0131c\u0131 grubunu hedefler ve bu nedenle belirli bir veri t\u00fcr\u00fcn\u00fc gerektirir. Bizim durumumuzda, hedef sosyal \u00f6\u011frenme platformlar\u0131 AKA’d\u0131r. Sonu\u00e7 olarak, AKA’ya benzer \u00f6\u011frenme platformlar\u0131ndan toplanan verileri bulmaya \u00e7al\u0131\u015ft\u0131k. AM\u0130\u00dcV ve OpenScout veri k\u00fcmelerini a\u015fa\u011f\u0131daki nedenlerden dolay\u0131 se\u00e7tik:<\/span><\/p>\n

              \n
            1. \n

              Veri k\u00fcmeleri, kullan\u0131c\u0131lar\u0131n \u00f6\u011frenme kaynaklar\u0131 hakk\u0131ndaki sosyal verilerini (derecelendirmeler, etiketler, incelemeler, vb.) sa\u011flar. Bu nedenle, veri k\u00fcmelerinin yap\u0131s\u0131, i\u00e7eri\u011fi ve hedef kullan\u0131c\u0131lar\u0131 AKA’n\u0131nkine benzerdir.<\/span><\/p>\n<\/li>\n

            2. \n

              Bu veri k\u00fcmeleri \u00fczerinde \u00f6neri sunan algoritmalar\u0131 kullanmak, AKA’n\u0131n ger\u00e7ek kullan\u0131c\u0131lar\u0131yla \u00e7evrimi\u00e7i olmadan \u00f6nce performanslar\u0131n\u0131 de\u011ferlendirmemize yard\u0131mc\u0131 olur.<\/span><\/p>\n<\/li>\n

            3. \n

              Hem AM\u0130\u00dcV hem de OpenScout veri k\u00fcmeleri, sosyal verilerin toplanmas\u0131 ve depolanmas\u0131 i\u00e7in standart bir \u00fcst veri belirtimi sunan BO\u00dcV (Ba\u011flamsal Otomatikle\u015ftirilmi\u015f \u00dcst Veri) format\u0131na (Schmitz vd., 2009) uygundur. BO\u00dcV ayr\u0131ca sosyal verilerin depolanmas\u0131 i\u00e7in AKA’da uygulanm\u0131\u015ft\u0131r.<\/span><\/p>\n<\/li>\n<\/ol>\n

              <\/a><\/a><\/a><\/a><\/a> Tablo 20.1. Se\u00e7ilen Veri K\u00fcmelerinin Detaylar\u0131<\/i><\/span><\/span><\/p>\n\n\n\n\n\n\n\n<\/colgroup>\n\n\n\n\n\n
              \n

              Veri k\u00fcmesi<\/span><\/p>\n<\/td>\n

              		\n

              Kullan\u0131c\u0131 say\u0131s\u0131<\/span><\/p>\n<\/td>\n

              		\n

              \u00d6ge say\u0131s\u0131<\/span><\/p>\n<\/td>\n

              		\n

              \u0130\u015flemler<\/span><\/p>\n<\/td>\n

              		\n

              Seyreklik (%)<\/span><\/p>\n<\/td>\n

              		\n

              Kaynak<\/span><\/p>\n<\/td>\n<\/tr>\n

              AM\u0130\u00dcV<\/span><\/span><\/td>\n631<\/span><\/span><\/td>\n12.571<\/span><\/span><\/td>\n23.032<\/span><\/span><\/td>\n99.70<\/span><\/span><\/td>\nAM\u0130\u00dcV portal\u0131<\/span><\/span><\/td>\n<\/tr>\n
              OpenScout<\/span><\/span><\/td>\n331<\/span><\/span><\/td>\n1.568<\/span><\/span><\/td>\n2.560<\/span><\/span><\/td>\n99.50<\/span><\/span><\/td>\nOpenScout portal\u0131<\/span><\/span><\/td>\n<\/tr>\n
              MovieLens<\/span><\/span><\/td>\n941<\/span><\/span><\/td>\n1.512<\/span><\/span><\/td>\n96.719<\/span><\/span><\/td>\n93.69<\/span><\/span><\/td>\nGroupLens ara\u015ft\u0131rmas\u0131<\/span><\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

              Bu iki veri k\u00fcmesinin yan\u0131 s\u0131ra, MovieLens veri k\u00fcmesini referans olarak test ettik, \u00e7\u00fcnk\u00fc bug\u00fcne kadar, e\u011fitim alan\u0131 genel olarak tavsiye sistemleri ile ilgilenen B\u0130MK TavSis konferans serisinin aksine, \u00e7al\u0131\u015fma i\u00e7in referans veri k\u00fcmeleri yetersiz kalm\u0131\u015ft\u0131r. Tablo 20.1, her \u00fc\u00e7 veri k\u00fcmesine genel bir bak\u0131\u015f sunmaktad\u0131r (Fazeli vd., 2014). E\u011fitsel veri k\u00fcmelerinden AM\u0130\u00dcV ve OpenScout\u2019un a\u015f\u0131r\u0131 derecede seyreklikten s\u0131k\u0131nt\u0131 ya\u015fad\u0131\u011f\u0131 dikkate al\u0131nmal\u0131d\u0131r. T\u00fcm verilere ait detaylar TD\u00d6AK 2014 makalemizde daha ayr\u0131nt\u0131l\u0131 olarak tan\u0131mlanmaktad\u0131r (Fazeli vd., 2014).<\/span><\/p>\n

              \u00c7evrimd\u0131\u015f\u0131 Veri \u00c7al\u0131\u015fmas\u0131<\/span><\/h2>\n

              Algoritmalar<\/span>. Bu ikinci ad\u0131mda, verilerimizle uyumlu \u00e7al\u0131\u015facak algoritmalar\u0131 se\u00e7meye \u00e7al\u0131\u015ft\u0131k. \u00d6ncelikle, \u00f6neri algoritmalar\u0131n\u0131 besleyecek olan giri\u015f verisini kontrol etmek \u00f6nemlidir. Bu durumda, AKA verileri, dolay\u0131s\u0131yla se\u00e7ilen veri k\u00fcmelerinin verilerini, kullan\u0131c\u0131lar\u0131n \u00f6\u011frenme kaynaklar\u0131 (\u00f6geler) ile etkile\u015fim verilerini i\u00e7erir. Bu nedenle, tavsiye sistemleri ailesinden \u0130\u015fbirlikli Filtreleme’den (\u0130F) faydalanmay\u0131 uygun g\u00f6rd\u00fck. \u0130F algoritmalar\u0131, i\u00e7erik esasl\u0131 tavsiye sistemleri taraf\u0131ndan kullan\u0131lan i\u00e7erik verilerinden daha ziyade kullan\u0131c\u0131lar\u0131n derecelendirmeler, yer imleri, g\u00f6r\u00fcn\u00fcmler, be\u011feniler, vb. gibi etkile\u015fim verilerine dayan\u0131r. \u0130F \u00f6nerileri \u201ct\u00fcr\u00fcne\u201d g\u00f6re bellek tabanl\u0131 veya model tabanl\u0131 olabilir; \u201cteknik\u201de at\u0131fta bulunarak madde baz\u0131nda veya kullan\u0131c\u0131 tabanl\u0131 olabilirler. Bu ayr\u0131mlar\u0131n ayr\u0131nt\u0131l\u0131 bir a\u00e7\u0131klamas\u0131 i\u00e7in B\u00f6l\u00fcm 4 Fazeli vd. (2014) bak\u0131n\u0131z. \u00c7al\u0131\u015fmam\u0131zda, t\u00fcm bellek t\u00fcrlerinden ve tekniklerinden faydaland\u0131k: bellek temelli, model temelli,kullan\u0131c\u0131 temelli ve \u00f6\u011fe temelli. \u015eekil 20.1, \u00fc\u00e7 temel ad\u0131mdan olu\u015fan deneysel y\u00f6ntemimizi g\u00f6stermektedir:<\/span><\/p>\n

               <\/p>\n

              \"\"<\/p>\n

              \u015eekil 20.1. Fazeli vd. (2014)’nin kulland\u0131klar\u0131 deneysel y\u00f6ntem.<\/i><\/span><\/span><\/p>\n

                \n
              1. \n

                Bellek bazl\u0131 \u0130F’lerin performans\u0131n\u0131 hem kullan\u0131c\u0131 hem de \u00fcr\u00fcn baz\u0131nda olmak \u00fczere farkl\u0131 benzerlik i\u015flevlerini kullanarak kar\u015f\u0131la\u015ft\u0131rd\u0131k.<\/span><\/p>\n<\/li>\n

              2. \n

                \u00d6rnek verilerimiz \u00fczerine en geli\u015fmi\u015f Matris \u00c7arpanlar\u0131na Ay\u0131rma y\u00f6ntemleri de d\u00e2hil olmak \u00fczere model tabanl\u0131 \u0130F’leri i\u015fledik.<\/span><\/p>\n<\/li>\n

              3. \n

                Ad\u0131m 1 ve 2’den en iyi performans g\u00f6steren algoritmalar\u0131n son bir kar\u015f\u0131la\u015ft\u0131rmas\u0131n\u0131 yapt\u0131k. Dayanaklara ek olarak, en yak\u0131n k kom\u015fusu (kNN) y\u00f6ntemini kullanarak kom\u015fu bulma mekanizmas\u0131n\u0131 geli\u015ftirmek i\u00e7in \u00f6nerilen grafik tabanl\u0131 bir yakla\u015f\u0131m\u0131 de\u011ferlendirdik (Fazeli vd., 2014).<\/span><\/p>\n<\/li>\n<\/ol>\n

                Performans De\u011ferlendirmesi. <\/span>Uygun veri k\u00fcmeleri ve \u00f6neri algoritmalar\u0131n\u0131 se\u00e7tikten sonra, aday algoritmalar\u0131n performans\u0131n\u0131 de\u011ferlendirme g\u00f6revine ula\u015f\u0131r\u0131z. Bunun i\u00e7in bir de\u011ferlendirme protokol\u00fc tan\u0131mlamam\u0131z gerekir (Herlocker vd. 2004). Bir de\u011ferlendirme protokol\u00fcn\u00fcn iyi bir a\u00e7\u0131klamas\u0131 a\u015fa\u011f\u0131daki sorular \u00fczerine e\u011filmelidir:<\/span><\/p>\n

                S1. Ne \u00f6l\u00e7\u00fclecek?<\/span><\/span><\/h4>\n

                Genel olarak, \u00e7o\u011fu \u00e7evrimd\u0131\u015f\u0131 tavsiye sistemi \u00e7al\u0131\u015fmalar\u0131nda, \u00fcretilen \u00f6nerilerin kestirim<\/i><\/span> do\u011frulu\u011funu \u00f6l\u00e7\u00fcyoruz. Bununla, bir e\u011fitim setini ve bir test setini kar\u015f\u0131la\u015ft\u0131rarak, derecelendirme kestirimlerinin ger\u00e7ek kestirimlerden ne kadar farkl\u0131 oldu\u011funu \u00f6l\u00e7mek istiyoruz. E\u011fitim ve test setleri, kullan\u0131c\u0131 derecelendirme verilerimizin (kullan\u0131c\u0131 etkile\u015fimi verileriyle ayn\u0131) b\u00f6l\u00fcnmesinden kaynaklanmaktad\u0131r. TD\u00d6AK 2014 \u00e7al\u0131\u015fmam\u0131zda, e\u011fitim seti ve test seti i\u00e7in kullan\u0131c\u0131 derecelendirmelerini s\u0131ras\u0131yla %80 ve %20’ye b\u00f6ld\u00fck. Bu t\u00fcr bir b\u00f6l\u00fcnme tavsiye sistem de\u011ferlendirmelerinde yayg\u0131n olarak kullan\u0131lmaktad\u0131r (Fazeli vd., 2014).<\/span><\/p>\n

                 <\/p>\n

                \"\"<\/p>\n

                \u015eekil 20.2. Grafik tabanl\u0131 \u0130F’nin F1’i ve kullan\u0131lan t\u00fcm veri k\u00fcmeleri i\u00e7in en iyi performans g\u00f6steren temel bellek tabanl\u0131 ve model tabanl\u0131 \u0130F’ler (Fazeli vd., 2014).<\/span><\/i><\/span><\/span><\/p>\n

                S2. Bir \u00f6neri sistemi \u00e7al\u0131\u015fmas\u0131 i\u00e7in uygun olan \u00f6l\u00e7\u00fcmler nelerdir?<\/span><\/span><\/h4>\n

                Girdi verilerimiz 5 y\u0131ld\u0131zl\u0131 derecelendirmeler gibi a\u00e7\u0131k kullan\u0131c\u0131 tercihleri \u200b\u200b<\/span>i\u00e7eriyorsa, MHO (mutlak hatalar ortalamas\u0131) veya KOKH (k\u00f6k ortalama kare hatas\u0131) kullanabiliriz. MHO ve KOKH’nin her ikisi de kullan\u0131c\u0131 derecelendirmeleriyle ayn\u0131 aral\u0131\u011f\u0131 takip eder; \u00f6rne\u011fin, e\u011fer veriler 5 y\u0131ld\u0131zl\u0131 derecelendirmeler i\u00e7eriyorsa, bu metrikler 1 ile 5 aras\u0131ndad\u0131r.<\/span><\/p>\n

                Girdi verileri g\u00f6r\u00fcn\u00fcmler, yer imleri, indirmeler vb. gibi kesin kullan\u0131c\u0131 tercihlerini i\u00e7eriyorsa Keskinlik, Hassasiyet ve F1 skorlar\u0131n\u0131 kullanabiliriz. F1 skorunu, \u00fcretilen \u00f6nerilerin do\u011frulu\u011funu ve kapsam\u0131n\u0131 de\u011ferlendirmede \u00f6nemli \u00f6l\u00e7\u00fctler olan keskinlik ve hassasiyeti birle\u015ftirdi\u011fi i\u00e7in kulland\u0131k (Herlocker vd., 2004). F1, 0 ile 1 aras\u0131ndad\u0131r.<\/span><\/p>\n

                Ek olarak n’ yi metrik olarak \u00f6l\u00e7\u00fclen, kesme<\/i><\/span> olarak da bilinen, en \u00fcstteki n tavsiyelerinde tan\u0131mlamam\u0131z gerekir.<\/span><\/p>\n

                Son olarak tan\u0131mlanan de\u011ferlendirme protokol\u00fcn\u00fc takip ederek veri k\u00fcmelerindeki aday algoritmalar\u0131n \u00e7al\u0131\u015ft\u0131r\u0131lmas\u0131yla ilgili sonu\u00e7lar\u0131 sunuyoruz. S\u0131n\u0131rl\u0131 alan nedeniyle sadece TD\u00d6AK 2014 makalemizin son sonu\u00e7lar\u0131n\u0131 burada sunuyoruz. Daha fazla sonu\u00e7 i\u00e7in l\u00fctfen orijinal makalenin 5.1 ve 5.2 b\u00f6l\u00fcmlerine (Fazeli vd., 2014) bak\u0131n\u0131z.<\/span><\/p>\n

                \u015eekil 20.2, en iyi performans g\u00f6steren bellek tabanl\u0131 \u0130F (Jaccard kNN), \u00e7izge tabanl\u0131 \u0130F ile kar\u015f\u0131la\u015ft\u0131r\u0131ld\u0131\u011f\u0131nda model tabanl\u0131 \u0130F (Bayes\u00e7i y\u00f6ntemi) in F1 sonu\u00e7lar\u0131n\u0131 g\u00f6stermektedir. X ekseni kullan\u0131lan veri k\u00fcmelerini ve y ekseni F1 de\u011ferlerini g\u00f6stermektedir. \u015eekil 20.2’nin g\u00f6sterdi\u011fi gibi \u00e7izge tabanl\u0131 yakla\u015f\u0131m AM\u0130\u00dcV ve MovieLens i\u00e7in en iyi performans\u0131 (%24) g\u00f6sterir ve se\u00e7ilen bellek -tabanl\u0131 ve model- tabanl\u0131 \u0130F’ler, \u00e7izge tabanl\u0131 \u0130F’den hem en sonra ikinci ve \u00fc\u00e7\u00fcnc\u00fc s\u0131rada yer almaktad\u0131r. OpenScout i\u00e7in, bellek tabanl\u0131 yakla\u015f\u0131m neredeyse %1 farkla daha iyi performans g\u00f6sterir.<\/span><\/p>\n

                Sonu\u00e7 olarak \u015fekil 20.2’de sunulan sonu\u00e7lara g\u00f6re \u00e7izge tabanl\u0131 yakla\u015f\u0131m\u0131n sosyal \u00f6\u011frenme platformlar\u0131 i\u00e7in etkili oldu\u011fu g\u00f6r\u00fclmektedir. Bu keskinlik ve yap\u0131lan \u00f6nerinin hassasiyetini etkili bir birle\u015fimi olan geli\u015fmi\u015f bir F1 ile yans\u0131t\u0131l\u0131r.<\/span><\/p>\n

                \u00d6neri Sisteminin Uygulanmas\u0131 ve Kullan\u0131c\u0131 \u00c7al\u0131\u015fmas\u0131<\/h3>\n

                E\u011fitim alan\u0131nda, kullan\u0131c\u0131 \u00e7al\u0131\u015fmalar\u0131n\u0131n \u00f6nemi her zamankinden daha da belirgin hale gelmi\u015ftir (Drachsler vd., 2015). E\u011fitimde tavsiye sistemlerinin temel amac\u0131 do\u011fru kestirimlerin \u00e7ok daha fazlas\u0131n\u0131 i\u00e7erdi\u011fi i\u00e7in fayda, yenilik ve \u00f6nerilerin \u00e7e\u015fitlili\u011fi gibi di\u011fer kalite g\u00f6stergelerini de yap\u0131land\u0131rmal\u0131d\u0131r. Bununla birlikte, tavsiye sistemi \u00e7al\u0131\u015fmalar\u0131n\u0131n \u00e7o\u011funlu\u011fu hala yaln\u0131zca \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmalar\u0131na dayanmaktad\u0131r. Bunun nedeni muhtemelen kullan\u0131c\u0131 \u00e7al\u0131\u015fmalar\u0131n\u0131n zaman al\u0131c\u0131 ve karma\u015f\u0131k olmas\u0131d\u0131r.<\/span><\/p>\n

                AKA verileri \u00fczerinde \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmas\u0131 yapt\u0131ktan sonra hedef platformumuzda bir kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131 y\u00fcr\u00fcten Fazeli vd. (2014) a\u00e7\u0131klad\u0131\u011f\u0131 \u00e7al\u0131\u015fmalar\u0131 daha da ileriye ta\u015f\u0131d\u0131k. Bunun i\u00e7in en iyi performans g\u00f6steren algoritmalar\u0131 AKA ile birle\u015ftirdik. Ger\u00e7ek AKA kullan\u0131c\u0131lar\u0131na, onlar i\u00e7in yapt\u0131\u011f\u0131m\u0131z \u00f6nerilerden memnun olup olmad\u0131klar\u0131n\u0131 sorduk. Bunun i\u00e7in \u015fu be\u015f fakt\u00f6r\u00fc kullanarak k\u0131sa bir anket haz\u0131rlad\u0131k: kullan\u0131\u015fl\u0131l\u0131k, do\u011fruluk, yenilik, \u00e7e\u015fitlilik ve tesad\u00fcfen de\u011ferli bir \u015feyler ke\u015ffetme yetene\u011fi. Bu veri \u00e7al\u0131\u015fmas\u0131n\u0131n ve takip eden kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131n\u0131n tam a\u00e7\u0131klamas\u0131 ve sonu\u00e7lar\u0131 hen\u00fcz yay\u0131nlanmam\u0131\u015ft\u0131r. Kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131, ger\u00e7ek AKA verileri \u00fczerinde y\u00fcr\u00fctt\u00fc\u011f\u00fcm\u00fcz veri \u00e7al\u0131\u015fmas\u0131n\u0131n sonu\u00e7lar\u0131n\u0131 do\u011frulamaz, tahmin do\u011frulu\u011fu gibi veri \u00e7al\u0131\u015fmalar\u0131n\u0131n ba\u015far\u0131 g\u00f6stergelerinin \u00f6tesine ge\u00e7ebilecek kullan\u0131c\u0131 \u00e7al\u0131\u015fmalar\u0131 yapman\u0131n olduk\u00e7a gerekli oldu\u011funu g\u00f6sterir.<\/span><\/p>\n

                Do\u011fruluk, tavsiye sistemlerini de\u011ferlendirmedeki \u00f6nemli \u00f6l\u00e7\u00fctlerden biridir ancak sadece bu metri\u011fe g\u00fcvenmek veri bilimcilerinin ve e\u011fitim teknologlar\u0131n\u0131n daha az etkili yollara y\u00f6nelmelerine neden olabilir.<\/span><\/p>\n

                PRAT\u0130K \u00c7IKARIMLAR VE SINIRLAMALAR<\/h2>\n

                E\u011fitsel veri k\u00fcmeleri, referans ba\u011flant\u0131lar\u0131 arac\u0131l\u0131\u011f\u0131yla halka a\u00e7\u0131k ve a\u00e7\u0131k\u00e7a kullan\u0131labilir olmad\u0131klar\u0131ndan e\u011fitim veri k\u00fcmelerinin \u00e7o\u011funa eri\u015fmek zordur. Dahas\u0131 ilgili \u00e7al\u0131\u015fmalara ait bulgular\u0131n kar\u015f\u0131la\u015ft\u0131r\u0131lmas\u0131 zordur, \u00f6rne\u011fin Verbert vd. (2011) ve Manouselis, Vuorikari, and Van Assche\u2019 nin (2010) \u00e7al\u0131\u015fmalar\u0131 gibi. Her ne kadar ayn\u0131 veri k\u00fcmelerini ve bu iki \u00e7al\u0131\u015fmada kullan\u0131lan baz\u0131 algoritmalar\u0131 uygulasak da \u00f6rnek \u00e7al\u0131\u015fmam\u0131z\u0131n sonu\u00e7lar\u0131 onlar\u0131n sonu\u00e7lar\u0131ndan farkl\u0131d\u0131r. Bu nedenle, \u00f6\u011frenme kaynaklar\u0131n\u0131n ki\u015fiselle\u015ftirilmesine ili\u015fkin kar\u015f\u0131la\u015ft\u0131rmalardan ek bilgi elde edemedik. Olas\u0131 sebeplerden biri \u00e7al\u0131\u015fmalar\u0131n ayn\u0131 veri k\u00fcmesinin farkl\u0131 versiyonlar\u0131n\u0131 kullanmas\u0131d\u0131r \u00e7\u00fcnk\u00fc toplanan veriler farkl\u0131 zamanlara aittir. \u00d6rne\u011fin AM\u0130\u00dcV veri k\u00fcmesi i\u00e7in farkl\u0131 s\u00fcr\u00fcmler mevcuttur. Asl\u0131nda, denemeler yapmak i\u00e7in ya da tavsiye sistemi toplulu\u011funda kar\u015f\u0131la\u015ft\u0131rma yapmak i\u00e7in benzersiz bir s\u00fcr\u00fcm belirlenmemi\u015ftir.<\/span><\/p>\n

                Bu sorun, maalesef e\u011fitim alan\u0131nda e-ticaret d\u00fcnyas\u0131nda bulunan MovieLens veri k\u00fcmesi1<\/a><\/sup> ile kar\u015f\u0131la\u015ft\u0131r\u0131labilecek alt\u0131n bir standart veri k\u00fcmesinin olmamas\u0131ndan kaynaklanmaktad\u0131r. Asl\u0131nda, \u00d6A toplulu\u011fu, farkl\u0131 ki\u015fiselle\u015ftirme yakla\u015f\u0131mlar\u0131 i\u00e7in ana referanslar seti olarak kullan\u0131labilecek birka\u00e7 temsili veri k\u00fcmesine ihtiya\u00e7 duymaktad\u0131r. Temel ama\u00e7, \u00d6A ara\u015ft\u0131rmas\u0131n\u0131 y\u00fcr\u00fctmek i\u00e7in standart bir veri format\u0131 elde etmektir. Bu fikir ba\u015flang\u0131\u00e7ta dataTEL projesi (Drachsler vd., 2011) taraf\u0131ndan \u00f6nerilmi\u015f ve daha sonra SoLAR \u00d6\u011frenme Analiti\u011fi Toplulu\u011fu taraf\u0131ndan takip edilmi\u015ftir (Ga\u0161evi\u0107 vd., 2011). KA\u00c7D’ler alan\u0131nda, Drachsler ve Kalz (2016), bu kar\u015f\u0131la\u015ft\u0131r\u0131labilir sonu\u00e7 eksikli\u011fini ve bilimsel sonu\u00e7lar\u0131 kar\u015f\u0131la\u015ft\u0131rmak i\u00e7in veri havuzlar\u0131n\u0131 kullanan bir ara\u015ft\u0131rma d\u00f6ng\u00fcs\u00fcne duyulan acil ihtiyac\u0131 tart\u0131\u015fm\u0131\u015flard\u0131r. Dahas\u0131, AB taraf\u0131ndan finanse edilen LinkedUp2<\/a><\/sup> adl\u0131 bir proje, ba\u011fl\u0131 veri kavramlar\u0131n\u0131 uygulayarak bir dizi alt\u0131n standart veri k\u00fcmesini sa\u011flama y\u00f6n\u00fcnde \u00fcmit verici bir yakla\u015f\u0131m izlemektedir (Bizer, Heath ve Berners \u2013 Lee, 2009). LinkedUp projesi, \u00f6\u011frenme analiti\u011fi ara\u015ft\u0131rmalar\u0131 i\u00e7in ba\u011flant\u0131l\u0131 bir veri havuzu sa\u011flamay\u0131 ve merkezi veri havuzu \u00fczerinden \u00e7e\u015fitli veri yar\u0131\u015fmalar\u0131n\u0131 y\u00fcr\u00fctmeyi ama\u00e7l\u0131yor.<\/span><\/p>\n

                Genel olarak, farkl\u0131 tavsiye sistemlerinin sonu\u00e7lar\u0131 veya e\u011fitim alan\u0131ndaki ki\u015fiselle\u015ftirme yakla\u015f\u0131mlar\u0131n\u0131n kar\u015f\u0131la\u015ft\u0131r\u0131lmas\u0131 algoritmalar\u0131n, \u00f6\u011frenen modellerinin, veri k\u00fcmelerinin ve de\u011ferlendirme \u00f6l\u00e7\u00fctlerinin \u00e7e\u015fitlili\u011fi nedeniyle hala g\u00fc\u00e7l\u00fckle yap\u0131labilmektedir(Drachsler vd., 2015; Manouselis vd., 2012).<\/span><\/p>\n

                SONU\u00c7<\/h2>\n

                Bu b\u00f6l\u00fcm\u00fcn temel amac\u0131 bir \u00f6\u011frenme ortam\u0131 i\u00e7in en uygun tavsiye sisteminin nas\u0131l tan\u0131mland\u0131\u011f\u0131n\u0131 g\u00f6stermektir. Bunu yapmak i\u00e7in, Drachsler vd.nin (2015) \u00f6\u011frenmede dan\u0131\u015fman sistemlerini de\u011ferlendirmek i\u00e7in sunduklar\u0131 standart metodolojiyi kullanarak bir \u00f6rnek veri \u00e7al\u0131\u015fmas\u0131n\u0131 takip ettik. Metodoloji d\u00f6rt temel basamaktan olu\u015fur:<\/span><\/p>\n

                  \n
                1. \n

                  Ger\u00e7ek verinin hen\u00fcz mevcut olmamas\u0131 durumu g\u00f6z \u00f6n\u00fcnde bulundurularak hedef verilerin benzeri olan ve tercihen de e\u011fitim alan\u0131ndan uygun veri k\u00fcmelerini se\u00e7in.<\/span><\/p>\n<\/li>\n

                2. \n

                  Girdi verilerine en uygun olan bir dizi aday \u00f6neri algoritmalar\u0131n\u0131 \u00e7al\u0131\u015ft\u0131r\u0131n. Bu basama\u011f\u0131n \u00e7\u0131kt\u0131s\u0131 girdi verileriyle en iyi \u015fekilde \u00e7al\u0131\u015facak \u00f6neri algoritmalar\u0131n\u0131 ortaya \u00e7\u0131karmal\u0131d\u0131r.<\/span><\/p>\n<\/li>\n

                3. \n

                  Kendileri i\u00e7in yap\u0131lan tavsiyelerde kullan\u0131c\u0131 memnuniyetini \u00f6l\u00e7mek amac\u0131yla bir kullan\u0131c\u0131 \u00e7al\u0131\u015fmas\u0131 ger\u00e7ekle\u015ftirin.<\/span><\/p>\n<\/li>\n

                4. \n

                  En iyi aday tavsiyeyi hedef \u00f6\u011frenme platformuna da\u011f\u0131t\u0131n.<\/span><\/p>\n<\/li>\n<\/ol>\n

                  Kullan\u0131c\u0131lara y\u00f6nelik yapt\u0131\u011f\u0131m\u0131z \u00e7al\u0131\u015fma sonu\u00e7lar\u0131n\u0131n, \u00e7evrimd\u0131\u015f\u0131 veri \u00e7al\u0131\u015fmas\u0131 sonu\u00e7lar\u0131n\u0131 do\u011frulamam\u0131\u015f olmas\u0131 ger\u00e7e\u011fi, her ne kadar zaman al\u0131c\u0131 ve karma\u015f\u0131k olsa da kullan\u0131c\u0131lara y\u00f6nelik \u00e7al\u0131\u015fmalar y\u00fcr\u00fct\u00fclmesinin \u00f6nemini ortaya koymaktad\u0131r.<\/span><\/p>\n

                  TE\u015eEKK\u00dcR B\u00d6L\u00dcM\u00dc<\/h2>\n

                  Bu b\u00f6l\u00fcm k\u0131smen AB 7. \u00c7er\u00e7eve Program\u0131 A\u00e7\u0131k Uzay Ara\u015ft\u0131rmalar\u0131 projesi kapsam\u0131nda finanse edilmi\u015ftir. Bu belge Avrupa Birli\u011fi’nin g\u00f6r\u00fc\u015flerini temsil etmemektedir ve Avrupa Birli\u011fi i\u00e7eri\u011finden kaynaklanabilecek herhangi bir kullan\u0131mdan sorumlu de\u011fildir. Hendrik Drachsler’in \u00e7al\u0131\u015fmalar\u0131 FP7 AB projesi \u00d6ATP taraf\u0131ndan desteklenmi\u015ftir.<\/span><\/p>\n

                  KAYNAK\u00c7A<\/h2>\n

                  Bizer, C., Heath, T., & Berners-Lee, T. (2009). Linked data: The story so far. International Journal on Semantic Web and Information Systems, 5<\/i>(3), 1\u201322. <\/span><\/span><\/p>\n

                  Dietze, S., Siemens, G., Taibi, D., & Drachsler, H. (2016). Editorial: Datasets for learning analytics. Journal of Learning Analytics, 3<\/i>(3), 307\u2013311. http:\/\/dx.doi.org\/10.18608\/jla.2016.32.15<\/span><\/span><\/p>\n

                  Drachsler, H., Hummel, H., van den Berg, B., Eshuis, J., Waterink, W., Nadolski, R., Berlanga, A., Boers, N., & Koper, R. (2009). Effects of the ISIS recommender system for navigation support in self-organised learning networks. Journal of Educational Technology & Society, 12<\/i>(3), 115\u2013126. <\/span><\/span><\/p>\n

                  Drachsler, H., & Kalz, M. (2016). The MOOC and learning analytics innovation cycle (MOLAC): A reflective summary of ongoing research and its challenges. Journal of Computer Assisted Learning, 32<\/i>(3), 281\u2013290. http:\/\/ doi.org\/ 10.1111\/jcal.12135 <\/span><\/span><\/p>\n

                  Drachsler, H., Rutledge, L., Van Rosmalen, P., Hummel, H., Pecceu, D., Arts, T., Hutten, E., & Koper, R. (2010). ReMashed: An usability study of a recommender system for mash-ups for learning. International Journal of Emerging Technologies in Learning, 5<\/i>. http:\/\/online-journals.org\/i-jet\/article\/view\/1191 <\/span><\/span><\/p>\n

                  Drachsler, H., Verbert, K., Santos, O., & Manouselis, N. (2015). Panorama of recommender systems to support learning. In F. Ricci, L. Rokach, & B. Shapira (Eds.), Recommender systems handbook<\/i>, 2nd ed. Springer. <\/span><\/span><\/p>\n

                  Drachsler, H., Verbert, K., Sicilia, M.-A., Wolpers, M., Manouselis, N., Vuorikari, R., Lindstaedt, S., & Fischer, F. (2011). dataTEL: Datasets for Technology Enhanced Learning \u2014 White Paper<\/i>. Stellar Open Archive. http: \/\/ dspace.ou.nl\/handle\/1820\/3846 <\/span><\/span><\/p>\n

                  Duval, E. (2011). Attention please! Learning analytics for visualization and recommendation. Proceedings of the 1st International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201911), 27 February\u20131 March 2011, Banff, AB, Canada (pp. 9\u201317). New York: ACM. <\/span><\/span><\/p>\n

                  Fazeli, S., Loni, B., Drachsler, H., & Sloep, P. (2014). Which recommender system can best fit social learning platforms? Proceedings of the 9th European Conference on Technology Enhanced Learning <\/i>(EC-TEL\u201914), 16\u201319 September 2014, Graz, Austria (pp. 84\u201397). <\/span><\/span><\/p>\n

                  Ga\u0161evi\u0107, G., Dawson, C., Ferguson, S. B., Duval, E., Verbert, K., & Baker, R. S. J. d. (2011, July 28). Open learning analytics: An integrated and modularized platform<\/i>. http:\/\/www.elearnspace.org\/blog\/wp-content\/uploads\/2016\/02\/ProposalLearningAnalyticsModel_SoLAR.pdf <\/span><\/span><\/p>\n

                  Greller, W., & Drachsler, H. (2012). Translating learning into numbers: A generic framework for learning analytics. Proceedings of the 2nd International Conference on Learning Analytics and Knowledge <\/i>(LAK\u201912), 29 April\u20132 May 2012, Vancouver, BC, Canada (pp. 42\u201357). New York: ACM. <\/span><\/span><\/p>\n

                  Herlocker, J. L., Konstan, J. A., Terveen, L. G., & Riedl, J. T. (2004). Evaluating collaborative filtering recommender systems. ACM Transactions on Information Systems, 22<\/i>(1), 5\u201353. <\/span><\/span><\/p>\n

                  Manouselis, N., Drachsler, H., Verbert, K., & Duval, E. (2012). Recommender systems for learning<\/i>. Springer Berlin Heidelberg. <\/span><\/span><\/p>\n

                  Manouselis, N., Vuorikari, R., & Van Assche, F. (2010). Collaborative recommendation of e-learning resources: An experimental investigation. Journal of Computer Assisted Learning, 26<\/i>(4), 227\u2013242. <\/span><\/span><\/p>\n

                  Pazzani, M. J., & Billsus, D. (2007). Content-based recommendation systems. In P. Brusilovsky, A. Kobsa, & W. Nejdl (Eds.), The adaptive web: Methods and strategies of web personalization <\/i>(pp. 325\u2013341). Lecture Notes in Computer Science vol. 4321. Springer. <\/span><\/span><\/p>\n

                  Schafer, J. B., Frankowski, D., Herlocker, J., & Sen, S. (2007). Collaborative filtering recommender systems. In P. Brusilovsky, A. Kobsa, & W. Nejdl (Eds.), The adaptive web: Methods and strategies of web personalization <\/i>(pp. 291\u2013324). Lecture Notes in Computer Science vol. 4321. Springer. <\/span><\/span><\/p>\n

                  Schmitz, H., Scheffel, M., Friedrich, M., Jahn, M., Niemann, K., Wolpers, M., & Augustin, S. (2009). CAMera for PLE. In U. Cress, V. Dimitrova, & M. Specht (Eds.), Learning in the Synergy of Multiple Disciplines: 4th European Conference on Technology Enhanced Learning (EC-TEL 2009) Nice, France, September 29\u2013October 2, 2009 Proceedings <\/i>(pp. 507\u2013520). Lecture Notes in Computer Science vol. 5794. Springer. <\/span><\/span><\/p>\n

                  Verbert, K., Drachsler, H., Manouselis, N., Wolpers, M., Vuorikari, R., & Duval, E. (2011). Dataset-driven research for improving recommender systems for learning. Proceedings of the 1st International Conference on Learning Analytics and Knowledge (LAK\u201911), 27 February\u20131 March 2011, Banff, AB, Canada (pp. 44\u201353). New York: ACM.<\/span><\/span><\/p>\n

                   <\/p>\n

                  \n
                  \n

                  1<\/a> https:\/\/epress.lib.uts.edu.au\/journals\/index.php\/JLA\/article\/view\/5071\/5600<\/span><\/p>\n<\/div>\n

                  \n

                  2<\/a> http:\/\/opendiscoveryspace.eu<\/span><\/p>\n<\/div>\n

                  \n

                  3<\/a> http:\/\/www.grouplens.org\/node\/73<\/span><\/p>\n<\/div>\n

                  \n

                  4<\/a> www.linkedup-project.eu<\/span><\/p>\n<\/div>\n","protected":false},"author":1,"menu_order":8,"template":"","meta":{"pb_show_title":"on","pb_short_title":"","pb_subtitle":"","pb_authors":[],"pb_section_license":""},"chapter-type":[48],"contributor":[],"license":[],"class_list":["post-104","chapter","type-chapter","status-publish","hentry","chapter-type-numberless"],"part":73,"_links":{"self":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/chapters\/104","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\/104\/revisions"}],"part":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/parts\/73"}],"metadata":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/chapters\/104\/metadata\/"}],"wp:attachment":[{"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/wp\/v2\/media?parent=104"}],"wp:term":[{"taxonomy":"chapter-type","embeddable":true,"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/pressbooks\/v2\/chapter-type?post=104"},{"taxonomy":"contributor","embeddable":true,"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/wp\/v2\/contributor?post=104"},{"taxonomy":"license","embeddable":true,"href":"https:\/\/acikkitap.com.tr\/oaek\/wp-json\/wp\/v2\/license?post=104"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}