Tez Türü: Doktora
Tezin Yürütüldüğü Kurum: Eskişehir Osmangazi Üniversitesi, Fen Bilimleri Enstitüsü, Türkiye
Tez Danışmanı: Veysel Yılmaz
Tezin Onay Tarihi: 2018
Tezin Dili: Türkçe
Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
Özet:
Nowadays, the volume of the available information is increasing day by day with the inclusion of internet in to our personal lives. However, it may be very difficult for a person to cluster big data and obtain/interpret useful information out of them. In order to reach more accurate and relevant information, recently the discussion of "Information Filtering Systems" has been started. The purpose of Information Filtering Systems is to provide the user with pure information from unwanted and cluttered data automatically. In this respect, "The Recommendation Systems" have been emerged. The main purpose of the recommendation systems is to estimate the likes of people in order to make recommendations accordingly. In this thesis, a recommendation system model is purposed to help the students, of Anadolu University Faculty of Economics and Administrative Science, in the selection of the courses to be taken in the future by predicting the grades of future compulsory and elective courses offered in students' curriculum. In the recommendation system model, the student-student and course-course similarities are used. In order to reflect the similarity, the well-known Pearson's correlation coefficient is used. Model and memory-based recommendation algorithms are used via similarity measure. In memory based collaborative filtering, product-based recommendation systems are used; In model based collaborative filtering, k-means, k nearest neighbours, and singular value decomposition is used to predict the students' grade scores. In the application part of the study, the results of four different algorithms are studied in detail. As a result, the singular value decomposition, because of its ability to show the latent pattern, is used as the prediction method for students. Among the recommendation algorithms used in this study, the minimum mean absolute error indicator is achieved by the singular value decomposition. Keywords: Recommendation Systems, Learning Analytics, Collaborative Filtering, Singular value decomposition (SVD), Clustering