Sentiment Classifier using Hugging Face Transformers

A modular NLP pipeline and interactive web app for sentiment analysis on text using Hugging Face Transformers. This project demonstrates scalable inference, binary and multi-class classification compatibility, model evaluation, and clean engineering practices.

Project Overview

• Goal: Predict whether a movie review expresses positive or negative sentiment
• Input Options: Upload your own CSV or use built-in sample data
• Models:
  • distilbert-base-uncased-finetuned-sst-2-english (binary)
  • tabularisai/multilingual-sentiment-analysis (multiclass)
• Interface: Built using Streamlit for a deployable, interactive UI
• Features:
  • Batch inference with tokenizer-aware chunking
  • Model selection and label mapping
  • Per-label filtering, explanation outputs, and downloadable results

Directory Structure

imdb-sentiment-classifier/
├── src/
│ ├── preprocess.py # Tokenization and decoding
│ ├── inference.py # Hugging Face model inference
├── app.py # Streamlit app
├── LICENSE # MIT License
├── README.md # You are here
└── requirements.txt # Project dependencies

App Demo

You can interact with the web app to upload review data, select the model type, and view predictions directly. Results can be filtered by label and downloaded as CSV.

Link to live app: https://sentiment-classifier-owen-wienczkowski.streamlit.app/

How to Run Locally

Make sure you're in the project root directory and have Python 3.8+.

1. Install dependencies

pip install -r requirements.txt

2. Run interactive web app

streamlit run app.py

Results for Web App (Example)

Skills Demonstrated

Hugging Face Transformers and inference pipelines

Streamlit interface design for real-time inference

Modular, production-style pipeline design

Tokenization, decoding, label mapping

Custom evaluation logic for multiclass-to-binary transitions

Multilingual model handling and class consolidation

Usable, downloadable web interface for non-technical users