Detrmining learner profiles and behavioral patterns in massive open online courses using clustering analysis: The case of Learn Turkish System

Within the scope of this thesis, the profile of learners was determined in a language MOOC (LMOOC) that allows people all over the world to learn the Turkish language, accordingly, it was determined whether learners form meaningful groups. At the same time, it was discovered to what extent learners' characteristics and learning behaviours explain their inclusion in a group. In this context, two different studies have been conducted. Within the scope of the first study, the demographic characteristics of the learners registered in the system, their information about their previous learning and their participation in the activities in the system were discovered using Exploratory Data Analysis (EDA) processes and the current status was revealed with the visualization technique. One of the interesting results achieved as a result of EDA, the results of more than half the learners in the system are Syrians and consist of men living in Turkey. More than half of the learners are at the undergraduate and above educational level as well as their online learning experience and their prior level in the Turkish language are low. When the learners' behavioural patterns on tutorials and drama videos investigated, it is seen that the tutorials consist of grammar rules are more preferred than the drama videos. Furthermore, most of the video watchers are residing in Turkey, and they are in the +51 age group. Video and visual activities were preferred more than the activities with text and keyboard input. In the second phase of the study, it has been investigated how the learners can be subgrouped due to their demographic (age, gender, employment status, education level), language-related (online course experience and Turkish language level) and online learning environment (watching videos, participating in short exams and participation in department activities) characteristics. At this stage, TwoStep clustering algorithm was used to include all categorical and continuous variables in the analysis. As a result of the analysis, five different subgroups emerged. These groups are named as (I) the highest engagement and diversity, (II) low engagement and non-online learning, (III) online learner and high engagement, (IV) adult and professional, (V) young and student. A two-stage approach has been adopted to validate the cluster analysis results. In the first stage, the resulting clusters were evaluated by controlling the Average Silhouette and Predictor Importance values. In the second stage, the discrimination of continuous and ordered variables that do not show normal distribution was checked by using non-parametric Kruskal-Wallis sequential one-way analysis of variance for Independent Groups. After the validation phase, it is seen that online course experience and education level are the variables with the highest discrimination. These two variables are followed by age, attendance to the quizzes, participation in the Listen and Choose the Visual activity, and the frequency of watching videos, respectively. This research process has enabled both to see the characteristics of learners in a Turkish LMOOC and whether their participation in the system differs according to these characteristics, and also to determine which variables are caused by these differences. In this way, the learner subsets were determined, and it is thought that a study was conducted to guide the practitioners in both the relevant literature and the processes of teaching Turkish as a foreign language.