Sentiment-Analysis-IMDB

Benchmark Results

Evaluating Fine Tuned RoBERTa's Embeddings via k-Means

Elbow method verifies the number of clusters should be 2

Problem Statement

This is a binary text classification problem where we predict the sentiment of movie reviews as either positive or negative. The classes are balanced.

XGBoost

It is trained on features aggregated from character-TFIDF and word-TFIDF. Character-TFIDF has been used to account for misspellings.

The XGBoost model minimizes a custom binary logistic objective and uses accuracy score as the evaluation metric.

The training phase includes validating the model to find the optimal number of boosting rounds with early stopping and sets the classification threshold to maximize the accuracy score on a validation set.

BERT, RoBERTa

These are pre-trained large language models that are fine-tuned by placing a classifier head on top.

Ensemble

This is an ensemble of XGBoost, BERT and RoBERTa based on majority voting.

Setup Instructions

Move into top-level directory

cd Sentiment-Analysis-IMDB

Install environment

conda env create -f environment.yml

Activate environment

conda activate sentiment-analysis

Install package

pip install -e src/sentiment-analysis

Including the optional -e flag will install sentiment-analysis in "editable" mode, meaning that instead of copying the files into your virtual environment, a symlink will be created to the files where they are.

Fetch data

python -m sentiment_analysis fetch

Download NLTK data

python -m nltk.downloader all

Run jupyter server

jupyter notebook notebooks/

You can now use the jupyter kernel to run notebooks.