Polarity Detection on Movie Review Dataset using CNN

In this notebook we use NLP techniques to clean and preprocess the data. After that, we build a CNN to predict polarity of each document.

Increase the data to get a better result

Dataset

Contents

Data Understanding

There are 2000 reviews belonging to 2 classes. Each document is a review and it can be negative or positive.

Maximum number of words for positive reviews is 1693 and for negative reviews is 1400. We need this information for text padding.

Data Cleaning

Here we load the documents and assign each one a target. We assign 0 to negative reviews and 1 to positive ones.

Other actions:

• Punctuation removal
• Stopwords removal
• Word tokenization
• Train-test split
  • 1600 documents for train and 400 documents for test

Data Preprocessing

• Tokenizing
• Encoding
• Text padding

Modeling

We use the following model