A Bayesian CNN-LSTM Model for Sentiment Analysis in Massive Open Online Courses MOOCs

Abstract

Massive Open Online Courses (MOOCs) are increasingly used by learn-ers to acquire knowledge and develop new skills. MOOCs provide a trove of data that can be leveraged to better assist learners, including behavioral data from built-in collaborative tools such as discussion boards and course wikis. Data tracing social interactions among learners are especially inter-esting as their analyses help improve MOOCs’ effectiveness. We particular-ly perform sentiment analysis on such data to predict learners at risk of dropping out, measure the success of the MOOC, and personalize the MOOC according to a learner’s behavior and detected emotions. In this pa-per, we propose a novel approach to sentiment analysis that combines the advantages of the deep learning architectures CNN and LSTM. To avoid highly uncertain predictions, we utilize a Bayesian neural network (BNN) model to quantify uncertainty within the sentiment analysis task. Our em-pirical results indicate that: 1) The Bayesian CNN-LSTM model provides interesting performance compared to other models (CNN-LSTM, CNN, LSTM) in terms of accuracy, precision, recall, and F1-Score; and 2) there is a high correlation between the sentiment in forum posts and the dropout rate in MOOCs.

Text Mining and Multi-Attribute Decision-Making-Based Course Improvement in Massive Open Online Courses

“…Kastrati et al [30] proposed a weakly supervised aspect-based sentiment analysis method for MOOC reviews, which can effectively reduce the dependence of the model on the number of training samples. Mrhar et al [31] proposed a sentiment analysis method that combines Bayesian neural networks, convolutional neural networks (CNNs), and long short-term memory (LSTM). This method demonstrates promising results in the sentiment analysis of MOOC comments.…”

New Deep Learning Model for Face Recognition and Registration in Distance Learning

“…At many levels of abstraction, the strength of a Convolution Neural Ne work (CNN) lies in obtaining a set of distinguishing feature maps. Convolutional neural networks have positively influenced the top of most other implementations [18]. In general, there are many CNN architectures, deep models such as GoogLeNet [19],ResNet [20] and [21].…”

Automated Reading Detection in an Online Exam

“…Convolutional neural networks were developed to extract spatial features of the data [30], [31]. In particular, they perform well dealing with tasks related to still images in Computer Vision [32] including object recognition and image classification. Despite various cutting-edge CNN-based models developed in CV, they are not well suited to problems of temporal nature.…”