Deep Learning Sentiment Analysis for MOOC Course Reviews

Abstract

MOOC is one of the most widely used online education platforms. Its online course reviewing is a channel for interaction between teachers and students. The emotions and attitudes reflected by the reviews are of great significance to teacher-student interaction. In order to better perform sentiment analysis on course reviews, this paper proposes a hybrid neural network model (BiAC-HNN) that comprises the Bi-LSTM, Attention and CNN. The model first converts the processed data set into a feature vector represented by a word vector through the word embedding layer, and then uses BiLSTM to obtain sentence context semantic information, merges Attention to obtain feature vectors with attention scores, and finally captures the correlation between contexts. It uses CNN to obtain higher-level key features, and inputs the features to the classification layer, and then uses softmax function to perform sentiment classification to get the sentiment tendency of the text. The experimental results, indicate comparing with Bi-LSTM, CNN and their enhanced models, BiAC-HNN has better accuracy and F1 value, which can improve the performance of text classification.