/*
Author:: Raj Mehrotra
Date:: 20-01-2019
*/
The famous Amazon fine food reviews dataset on Kaggle for text classification.
First the data is cleaned and pre-processed using standard NLP techniques like tokenization,stemming ,stop-words removal among others. Then I have performed sentiment analysis on the dataset using different approaches.
1) The first approach used is the traditional Bag-of-Words (BOW) approach using Tfidf and Count Vectorizer in scikit. In this case the word level tfidf vectorizer using the Logistic Regression gave best results for me with an accuracy of about a little more than 86%.
2) In the second approach I have used the Google Word2Vec word embeddings from Gensim librray and trained a LSTM model in Keras on them. This method also gives an accuracy of about 85%.