Sentiment Analysis of IMDB Movie Reviews

This program trains a model which is used to predict the sentiment of IMDB movie reviews. The data set we have used can be found at https://ai.stanford.edu/~amaas/data/sentiment/. The model we have trained is capable of achieving an accuracy of approximately >80%.

Extracting the Data

The .tar file has been included in the ./data directory but can alternatively be downloaded from here. The data should be extracted from the .tar file and should result in the following directory structure:

├── data
│   ├── aclImdb      <==== Extracted Folder
│   │   ├── test
│   │   ├── train
│   │   ├── imdb.vocab
│   │   ├── imdbEr.txt
│   │   └── README
│   └── aclImdb_v1.tar.gz
├── models
├── utils
├── README.MD
├── test_NLP.py
└── train_NLP.py

Once the data has been extracted we can begin training or testing the models.

NLTK Dependencies

It is important to note that before training or testing we also need to download NLTK modules. This can be done by running the following python file ./utils/nltk_packages.py.

Training Models

Models can be trained by running train_NLP.py. The model will then be saved in the ./models directory as NLP_Model.h5.

Testing Models

The model saved as ./models/NLP_Model.h5 can be tested by running test_NLP.py.