Sentiment Polarity Classification of Retail Product Reviews Using Machine Learning and Deep Learning

No Thumbnail Available
Kalita, Aneesh
Issue Date
MSc in Data Analytics
Dublin Business School
Items in eSource are protected by copyright. Previously published items are made available in accordance with the copyright policy of the publisher/copyright holder.
A comparative study has been undertaken to compare the performance of conventional machine learning classifiers and deep learning models to classify the sentiment polarity of the Amazon musical instrument reviews dataset. Text preprocessing and the Tf-idf feature vectorisation techniques were applied to the machine learning models. The 10-fold cross validation technique and Grid-search hyperparameter tuning were implemented to compare the performance of the Multinomial Naïve Bayes classifier, Logistic regression classifier and the Linear Support Vector classifier. The Google pre-trained Word2vec word embeddings and a custom word embedding trained using the Word2vec algorithm on the Amazon reviews dataset were used to train the artificial neural network (ANN) and recurrent neural network long short-term neural network (RNN-LSTM) models. The RNN-LSTM trained using the custom-trained word embeddings achieved the best accuracy and F1-score of 92.31% followed by the RNN-LSTM trained using the Google word2vec embeddings with an accuracy and F1-score of 91.92%. The SVM classifier achieved the greatest accuracy of 91.82% among the machine learning classifiers.