IMDb Sentiment Classification

Sentiment classification project on the IMDb movie review dataset with a classical baseline and a transformer-based classifier.

Overview

This project studies binary sentiment classification on IMDb reviews. The workflow includes dataset loading, exploratory analysis, a TF-IDF plus logistic regression baseline, a BERT-based classifier, and artifact generation for model comparison and error analysis.

Main Setup

• Dataset: imdb from Hugging Face Datasets
• Task: binary sentiment classification
• Baseline: TF-IDF + Logistic Regression
• Main model: BERT sequence classification

Results Snapshot

• Baseline accuracy: 0.8831
• BERT evaluation accuracy: 0.8960
• BERT evaluation F1: 0.8998
• Improvement over baseline: 0.0129

Repository Contents

• nlp-2.ipynb: main notebook covering EDA, baseline modeling, transformer training, and evaluation
• artifacts/baseline_results.json: baseline metrics
• artifacts/baseline_classification_report.txt: baseline classification report
• artifacts/baseline_wrong_predictions.csv: baseline error examples
• artifacts/bert_eval_results.json: transformer evaluation metrics
• artifacts/bert_summary.json: summary of the BERT run
• artifacts/bert_classification_report.txt: transformer classification report
• artifacts/bert_wrong_predictions.csv: transformer error examples
• artifacts/model_comparison.csv: comparison between the baseline and transformer setups

How To Use

1. Create a Python environment.
2. Install the required dependencies such as datasets, transformers, accelerate, evaluate, scikit-learn, pandas, numpy, matplotlib, seaborn, and nltk.
3. Open nlp-2.ipynb in Jupyter.
4. Run the notebook in order to reproduce dataset loading, baseline training, BERT fine-tuning, and evaluation.

Project Focus

• sentiment analysis
• IMDb reviews
• classical vs transformer comparison
• error analysis
• reproducible evaluation artifacts