Sentiment Analysis on Massive Open Online Courses (MOOCs): Multi-Factor Analysis, and Machine Learning Approach

Introduction

E-learning is the process of learning, educating, or training using digital or electronic based technologies. E-learning permits learners to build up and develop their knowledge independently of time and place. Recently, Massive Open Online Courses (MOOCs) are changing the way of online learning. MOOCs break the limitations of traditional online courses as they provide more flexibility in terms of when and where to take the courses through many features and components. MOOCs platforms generally use video lectures, reading texts, online assessments, quizzes and collaborative projects (Pursel et al. 2016; Pappano 2012). Moreover, MOOCs are usually supported by discussion forums to reinforce the interactions between different learning process actors.

Although the many advantages and available features that MOOCs provide, they encounter very serious problems such as the high dropout rate (Parr 2013). For instance, Amnueypornsakul et al. (2014) declare that MOOCs have less than 13% as a passing rate. Similarly, Breslow et al. (2013) determine a completion rate of 15% in MOOCs. In addition, Ho et al. (2014) affirm that only 5% of learners received the course certificate. Reich & Ruipérez-Valiente (2019), confirm a retention rate of 7% in the 2016–2017 cohort on all MOOCs taught on edX. Also, studies prove that MOOCs lack interaction between learners and instructors comparing to other educational approaches such as blended learning (Jia et al. 2019). Most students who dropout are not able to engage determinedly in the course materials, as many learners endure to stay cognitively and emotionally engaged. One way to increase learners-instructors interactions and to supervise learners’ progression during courseware is to analyse discussion forums. Discussion forums are a crucial part of the learning process, they provide students and instructors with an interactive collaborative environment for exchanging ideas and sharing opinions (Andresen 2009). Learners in the discussion forums come from different backgrounds; sharing textually their ideas, thoughts, feelings, knowledge and struggles toward different learning objects. Discussion forums are free open platforms where learners can explicitly express whatever cross their minds. Discussion forums open a great opportunity for tutors and the course’s instructors to analyse those textual data and extract knowledge about each individual learner during the learning process.

Sentiments are thoughts, attitudes, or mental perceptions. Sentiment analysis is commonly used as a general term related to extracting subjective information related to human opinions and emotions from texts (Pang & Lee 2008; Gilbert & Hutto 2014). Sentiment analysis is also associated with different research fields such as natural language processing, machine learning and data mining (Hailong et al. 2014). The sentiment is widely associated with affections, emotions, and feelings, and it profoundly affects learning. It may be defined as a determined opinion reflective of one’s feelings (Pang & Lee 2008). The use of sentiment analysis can shed the light on relevant knowledge obtained from unstructured text, which can be useful for decision-making (Phan et al. 2019). In the education domain, learning is not a cold mental activity. It is interspersed with different kind of emotions and sentiments. Sentiment analysis can show an individual’s inner cognitive development (Imai 2010), analyse students’ feedback in real-time (Altrabsheh et al. 2013), affect students’ subjective perceptions and judgement, individually (Molinari et al. 2013) or in collaborative groups (Zheng & Huang 2016). There are three techniques for conducting a sentiment analysis study; lexicon-based approach, machine learning-based approach and hybrid. The lexicon-based approach is based on creating a manual sentiment lexicon dictionary (Ahire 2014). As for the machine learning-based approach, it uses supervised machine learning algorithms and linguistic features (Medhat et al. 2014). The hybrid approach is a combination of both lexicon-based approaches and machine learning approach.