Deep Learning with Word Embedding Modeling for a Sentiment Analysis of Online Reviews

Abstract

Recently, online buyers have been able to express their opinions about a variety of products, restaurants and services by writing online reviews. Opinions have subsequently become a new, important, and influential source of information for decision-making. This paper implements binary and multiclass sentiment classifications on a dataset of online reviews. The experiments are performed using restaurant reviews from Yelp to predict ratings based on the content of the reviews. This paper investigates various structures of neural networks on restaurant reviews, such as recurrent neural networks (RNNs) with long short-term memory (LSTM), RNNs with bidirectional LSTM (Bi-LSTM) and convolutional neural networks (CNNs). The reviews were first converted into vectors during preprocessing using various features: pretrained word2vec and global vector (GloVe) embedding. The efficacy of these text classification techniques was examined and compared. The classification performance was evaluated using different metrics: the accuracy, confusion matrix, recall, specificity, precision, F1 score, receiver-operating characteristic (ROC) curve, and the area under the curve (AUC). The results showed that the RNN model achieved a better accuracy score with Bi-LSTM for both binary and multiple sentiment classification tasks.