Sentiment classification of Amazon fine food reviews

A sentiment analysis implementation on the Amazon fine food reviews dataset.

Overview

In this implementation, i used pretrained word embeddings GloVe 100d as the input of my deep learning model using Bidirectional LSTM.

Created by Ngo Quang Huy @ngoquanghuy99

Email: ngoquanghuy1999lp@gmail.com

Results on testing set

Download model

label	precision	recall	f1-score	support
0	0.79	0.73	0.76	11394
1	0.95	0.96	0.96	61388
accuracy			0.93	72782

Getting started

Install dependencies

Requirements

• tensorflow>=2.3.1

• keras>=2.4.3

  $ pip3 install -r requirements.txt
  

Training

$ python train.py

Accuracy graph:

Loss graph:

Fine tuning model by changing hyperparameters in config.py

Testing

On the assumption that you want to predict directly

$ python test.py --review "this is the worst thing i\'ve ever bought"

Output:

$ 'this is the worst thing i've ever bought' is NEGATIVE 82.63011127710342%