Assignment - Sentiment Analysis of Movie Reviews

Problem Statement: Apply the bag of words technique to train ML models for sentiment analysis using data from the "Sentiment Analysis on Movie Reviews" Kaggle competition.

Outline:

1. Download and Explore Dataset
2. Implement the Bag of Words Technique
3. Train baseline model & submit to Kaggle
4. Train & finetune different ML models
5. Document & submit your notebook

Dataset: https://www.kaggle.com/c/sentiment-analysis-on-movie-reviews

Download and Explore the Data

Outline:

1. Download Dataset from Kaggle
2. Explore and visualize data

Download Dataset from Kaggle

• Read the "Description", "Evaluation" and "Data" sections on the Kaggle competition page carefully
• Make sure to download the kaggle.json file from your Kaggle account and upload it on Colab

Explore and Visualize Data

• Load the train, test, and submission files using Pandas
• Explore rows, columns, sample values etc.
• Visualize distribution of target columns

Summarize your insights and learnings from the dataset below:

• ???
• ???
• ???

Implement Bag of Words Model

Outline:

1. Learn the vocabulary using CountVectorizer
2. Transform training and test data

Learn Vocabulary using CountVectorizer

• Create custom tokenizer with stemming
• Create a list of stop words
• Configure and create CountVectorizer
• Learn vocubulary from training set
• View sample entries from vocabulary

Transform Training & Test Data

• Transform phrases from training set
• Transform phrases from test set
• Look at some example values

Train Baseline Model & Submit to Kaggle

1. Split training and validation sets
2. Train logistic regression model
3. Study predictions on sample phrases
4. Make predictions and submit to Kaggle

Split Training and Validation Sets

Tip: Don't use a random sample for validation set (why?)

Train Logistic Regression Model

Study Predictions on Sample Inputs

Make Predictions & Submit to Kaggle

1. Make predictions on Test Dataset
2. Generate & submit CSV on Kaggle
3. Add screenshot of your score

Train & Finetune Different ML Models

Train & finetune at least 2 other types of models

Models to try:

• Decision Trees
• Random Forests
• Gradient Boosting
• Naive Bayes
• SVM

Optional:

• Use PCA for dimensionality reduction
• Compare classification vs regression models

Tips:

• If using a regression model, make sure to round predictions to integer and clip to the range [1,5]
• Track your progress in a copy of this experiment tracking spreadsheet

Model 1

Model 2

Model 3

Best Model:

???

(include Kaggle score screenshot)

Submission and Future Work

How to make a submission:

• Add documentation and explanations
• Clean up any stray code/comments
• Include a screenshot of your best score
• Make a submission on the assignment page
• Review evaluation criteria carefully

Future work:

• Try more machine learning models
• Try configuring CountVectorizer differently
• Try approaches other than bag of words