🎤 TED Talks Recommendation System

Content-based recommender that suggests relevant TED Talks based on your interests using TF-IDF + Cosine Similarity + Pearson Correlation.

---

✨ Features

• Clean modular project structure (not a single Jupyter notebook)
• Text preprocessing (stopwords removal + punctuation cleaning)
• TF-IDF vectorization
• Dual similarity scoring:
  • Cosine similarity (angle-based)
  • Pearson correlation (linear relationship)
• Combined ranking of top-N most relevant talks
• Easy to extend (add new similarity metrics, evaluation, UI, etc.)

🖼️ Example Output

Query:
"Climate change and impact on health and carbon footprint"

Recommended talks (example):

main_speaker	details
Al Gore	... climate change health impacts carbon emissions ...
Johan Rockström	... planetary boundaries climate health connection ...
Christiana Figueres	... Paris agreement health co-benefits ...

📂 Project Structure

ted-talks-recommender/
├── data/
│   └── tedx_dataset.csv               # original dataset
├── src/
│   ├── init.py
│   ├── preprocessing.py               # data loading & text cleaning
│   ├── model.py                       # TF-IDF + similarity computation
│   └── utils.py                       # helper functions
├── recommend.py                       # main recommendation script
├── README.md
└── requirements.txt                   (recommended)

🚀 Quick Start

1. Prerequisites

• Python 3.8+
• pandas, scikit-learn, nltk, scipy

2. Installation

# Clone the repository
git clone https://github.com/YOUR-USERNAME/ted-talks-recommender.git
cd ted-talks-recommender

# (Recommended) Create virtual environment
python -m venv venv
source venv/bin/activate    # Linux / macOS
venv\Scripts\activate       # Windows

# Install dependencies
pip install pandas scikit-learn nltk scipy

3. Download NLTK data (only needed once)

import nltk
nltk.download('stopwords')

4. Run the recommender

python recommend.py

Or modify the query directly in recommend.py:

query = [
    "Climate change and impact on health and carbon footprint",
    # "machine learning ethics bias fairness",
    # "future of education technology children",
]

🛠️ How It Works (Pipeline)

Load & Preprocess

• Read CSV → keep main_speaker, title, details
• Merge title + details
• Lowercase → remove stopwords → remove punctuation

Vectorization

• TfidfVectorizer on cleaned details column

Similarity Calculation

• Transform user query to TF-IDF vector
• Compute cosine similarity for each document
• Compute Pearson correlation for each document
• Sort primarily by cosine, secondarily by Pearson
• Return top-N results

🔧 Possible Improvements (Future Work)

• Add more preprocessing (lemmatization, stemming)
• Try sentence transformers / BERT embeddings
• Add evaluation metrics (precision@K, NDCG, user study)
• Create a simple Streamlit / Gradio web interface
• Support multi-query / query expansion
• Add speaker / event / year filters

📄 License

MIT License
Feel free to use this code for learning, personal projects, or portfolios.
Made with ❤️ by Youssef Mohammed
Happy recommending! 🎬