Sentiment analysis with Multinomial Naive Bayes Classifier

This was the first project in Machine Learning course on Faculty Of Computer Science.

Problem description

The project included several parts. This repository contains solutions for each assignment.

• Linear regression - 2a.py and 2b.py
• Instance-based models (KNN on Iris dataset) - 3a.py, 3b.py and 3c.py
• Multinomial Naive Bayes Classifier (Sentiment analysis of IMDB movie reviews) - 4.py

A detailed explanation is in statement.pdf

Dataset

Dataset included 2500 IMDB movie reviews, 1250 negative and 1250 positive.

Solution

For the first two problems I used TensorFlow, and the fourth problem was done in pure python while text processing was done with nltk library. The solution for the fourth problem was fairly simple, I used BoW histogram for feature extraction and on top of that straight forward Multinomial Naive Bayes. This method produces an accuracy of 90 - 94 % on the training set and 80 - 87 % on the test set.

Prerequisites

• TensorFlow
• SciPy
• nltk