Article Preview
TopIntroduction
With the emergence of educational big data, several novel theories and methods have been successively proposed in the field of learning analytics (Dringus & Ellis, 2005; Miyoshi & Nakagami, 2007; Popescu & Etzioni, 2005). In educational text mining, the latent Dirichlet allocation (LDA) method has been developed to identify key topics discussed by students. This method is a mathematical model that automatically classifies massive amounts of text and labels it by topic. Within intelligent information processing, many studies have focused on the topic model. Most of them concern the tuning of model parameters, and they have applied this method to the mining of users’ opinions in business contexts, but few relevant studies have been performed in education. In online education, the textual data in massive open online courses (MOOCs) offers valuable opportunities for understanding feedback from students. However, educators and administrators cannot generally observe individual learning process in any practical way (Moghaddam & Ester,2010; Ramesh, Goldwasser & Huang,2013; Ravichandran, Kulanthaivel & Chellatamilan, 2015). In this vein, LDA is considered an important tool for assessing the opinions and interests of students in discussion settings. The purpose of the present work is to improve LDA models and so adapt educational textual data mining. The improved LDA can be used to extract topics of interest from individuals. An LDA model can be adjusted to capture the topics and the word sequences characterizing each topic from a personal perspective. Topics for each learner participating in online courses can be determined, and the likelihood that a given learner was interested in any of these topics would be computed accurately. A great deal of information can be derived by mining learners’ reviews and some probability graphics can be produced to express the hottest topics associated with this course or the technical problem topic for educators and administrators. Educators can easily observe the learning process and identify problems specific to each student.
This paper aims to mine online learners’ topics of interest. Unlike existing models, this one creates a map for the author-documented distribution based on the fact that each author may produce many documents; meaning that each learner is corresponding to their various course reviews. Based on this idea, the document-topic relationship can be transformed into the author-topic relationship by the variant of LDA, here called the author topic model (ATM). The topics of interest associated with each author can be mined indirectly. The study is organized as follows. Section 2 provides a literature review of related works. Section 3 presents the improved author topic model and describes the design of the algorithm. Section 4 describes the experimental processes, results, and analysis. Section 5 presents the conclusion, discusses the implications of the present work, and suggests directions for further study.