The topic of my master thesis is sentiment analysis of Amazon reviews. I will explore different approaches incl. traditional machine learning method and neural networks for this task.
In traditional method, Bag of words and Tf-idf features are used for Naive Bayes and Logistic Regression model.
In deep learning NLP, firstly use pre-trained word2vec embeddings and LSTM to train. And second approach is to use pre-trained language models trained on large text corpus and fine-tuning on Amazon dataset like ULMFiT.