# EcoTweet Sentiment Analysis with BERT

This project applies **Natural Language Processing (NLP)** techniques to perform **sentiment analysis** on tweets discussing **eco-friendly and sustainable products**, such as biodegradable packaging, reusable bottles, and compostable materials.

I use **BERT (Bidirectional Encoder Representations from Transformers)** — a deep learning model by Google that understands context in both directions — and fine-tune it specifically on tweets about sustainability. The goal is to predict whether a tweet expresses a **positive** or **negative** sentiment about a green product.

This analysis helps in identifying **consumer attitudes**, tracking **greenwashing**, and informing **sustainability-focused business and policy decisions**.
 



##  Project Objective

To build and fine-tune a BERT-based sentiment classifier that can detect **positive** or **negative** sentiments in eco-related tweets. This helps in identifying public perception, detecting green washing, and supporting sustainable product development.

---

##  Project Structure

eco_tweet_sentiment/
│
├── data/    # Raw and synthetic tweet datasets
├── models/    # Fine-tuned BERT model outputs
├── notebooks/    # Main Jupyter/Colab notebook
├── venv/          # Python virtual environment
├── .gitignore     # Ignore model checkpoints, cache, etc.
├── README.md      # Project description and instructions
└── requirements.txt    # Python package dependencies


##  Dataset

We use two types of tweet data:
1. **Pre-labeled tweets** from `nltk.twitter_samples`:
   - 5,000 positive tweets
   - 5,000 negative tweets

2. **Synthetic eco-tweets**:
   - 500 positive eco tweets (generated)
   - 500 negative eco tweets (generated)

---

##  Methods & Tools

| Step | Description |
|------|-------------|
| **Data Loading** | Load NLTK twitter samples |
| **Text Preprocessing** | Lowercasing, tokenization, padding |
| **Model** | `bert-base-uncased` from Hugging Face |
| **Fine-Tuning** | BERT fine-tuned on synthetic eco tweets |
| **Evaluation** | Accuracy, loss, predictions on custom tweets |

---

##  Libraries Used

- [`transformers`](https://huggingface.co/transformers/)
- [`torch`](https://pytorch.org/)
- [`nltk`](https://www.nltk.org/)
- [`scikit-learn`](https://scikit-learn.org/)
- `pandas`, `matplotlib`
- `wandb` (for experiment tracking, optional)


author: Henok Yoseph
Email: henokapril@gmail.com 
github: https://github.com/aprilyab