Comment Data Mining to Estimate Student Performance Considering Consecutive Lessons.

Abstract

Introduction Nowadays, higher education institutions play a crucial role in improving the quality and the efficacy of education. Thus, it is important to determine methods that can provide quality education to students. One way to enhance the quality of educational processes is to focus on learners' activities and attitudes, which are currently unknown to the decision makers and very valuable in analyzing students' behavior, predicting performance and assisting courseware authors in detecting shortcomings and providing feedback to improve student performance. However, it is a difficult task to predict students' performance by analysing their behavior because each student has his/her own characteristics that influence their performance, such as their demographic, cultural, social, or psychological profile, previous schooling, prior academic performance, interactions between students and faculty, etc. (Yadav & Pal, 2012; Chalaris, Gritzalis, Maragoudakis, Sgouropoulou, & Tsolakidis, 2014; Natek & Zwilling, 2014). Comment data can potentially eliminate barriers between students and their teacher, and provide opportunities so that students can think back on their learning behaviors in connection with the subject. Thus, it becomes an essential accessory to support their learning. Further, comment data enables students to express themselves, their attitudes, and interactions, and to reflect their learning activities and difficulties for each lesson, especially for those having introverted characters and who are not comfortable expressing their views or asking questions. In addition, it allows teachers at the same time to improve their way of giving lessons, and of contacting students. In this paper, we synthesize the learning analytics (LA) and data mining (DM) approaches to explore the development of more usable prediction models of student performance using free-style comment data written by students after every lesson. The ability to predict the final performance of a student has been growing steadily in education. There are specific student characteristics which can be associated with the student success rate. One of the practical applications of student performance prediction is to find relationships between the work done by each student and the precise mark obtained. If we could predict student performance in advance, a feedback process could help to improve the learning process of the students during the course (Natek & Zwilling, 2014; Zafra & Ventura, 2009; Xing, Guo, Petakovic, & Goggins, 2015). To further contribute to the understanding of individual students in the class, this paper presents a study that examines different types of comment data (i) to find students' situations, tendencies and attitudes by predicting final student grades, (ii) to achieve further improvement in predicting final student grades by considering the student grades predicted in consecutive lessons and to keep track of the student learning situation, and (iii) to examine the characteristics of comment data collected from three different classes affecting the accuracy of final student grade prediction. The objective of our study is to find a reply to the following research questions: Question 1. Which topic model has the best results in predicting student performance, pLSA or LDA? Question 2. What comment format is the best predictor? Question 3. Is it possible to make an accurate prediction in early lessons, such as by lesson five? Question 4. Is there any relation between the difficulty of a subject and accuracy in predicting final student grades? The rest of the paper is organized as follows. Related work gives an overview of some related literature. Preliminary introduces an overview of our research and the procedures of the proposed methods. Methodology describes our proposed methods. Results discusses some of the highlighted experimental results. …