Automated Sentiment Analysis of Movie Reviews using various approaches including sklearn models, keras models & transfer learning

The goal for this analysis is to predict if a review rates the movie positively or negatively. Inside this dataset, there are 25,000 labelled movies reviews for training, 50,000 unlabeled reviews for training, and 25,000 reviews for testing.

• IMDB movie reviews dataset
• http://ai.stanford.edu/~amaas/data/sentiment
• https://www.kaggle.com/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews
• Contains 25000 positive and 25000 negative reviews
• Contains at most reviews per movie
• At least 7 stars out of 10 → positive (label = 1)
• At most 4 stars out of 10 → negative (label = 0)

Notebooks

1. Exploration and Preprocessing
2. Base Models (Logistic Regression, Multinomial NB)
3. Keras Models
4. PyTorch RNN Model
5. BERT Fine Tuned Model

Dataset

• The data used for this problem can be found at: https://www.kaggle.com/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews
• The preprocessed data can be found at : https://drive.google.com/file/d/1-KrTwLg3b2NcHFafK_lrKeyTvUK3NhiL/view?usp=sharing

Accuracies Achieved:

• Logistic Regression | 90.79 %
• Support Vector Machine | 91.08 %
• Multinomial Naive Bayes | 91.32 %
• Simple Neural Net Keras | 92.83 %
• RNN LSTM PyTorch | 86.04 %
• BERT Fine Tuning | 91.68 %

WordClouds

Positive Reviews WordCloud

Negative Reviews WordCloud