Recommender Systems From Scratch

This repository implements foundational and intermediate recommender system models entirely from scratch, without relying on external recommender libraries such as Surprise or implicit. The goal is to build a deep, hands-on understanding of how core recommendation algorithms work under the hood.

---

Project Structure

• notebooks/

  • RS_Chapter1_Foundations.ipynb
    Exploratory data analysis and regression-based prediction models.
  • RS_Chapter2_Collaborative_Filtering.ipynb
    User-based and item-based collaborative filtering implementations.
  • RS_Chapter3_Content_Based.ipynb
    Personalized and non-personalized content-based recommendation models.

• src/

  • Helper functions such as KNN-based similarity search and evaluation metrics.

• data/

  • Joke ratings, text datasets, and stopword lists used for building models.

---

Setup Instructions

Clone the repository and install the required dependencies:

git clone https://github.com/wins-wang/recommender-systems-from-scratch.git
cd recommender-systems-from-scratch
pip install -r requirements.txt

---

Covered Topics

Chapter 1: Foundations

• Exploratory Data Analysis (EDA) and visualization
• Simple and Ridge Regression for rating prediction
• Popularity-based baseline recommender

Chapter 2: Collaborative Filtering

• User-User similarity computation (Pearson correlation)
• Item-Item similarity computation (Cosine similarity)
• K-Nearest Neighbors (KNN) prediction for rating estimation
• Top-N recommendations based on collaborative filtering
• Evaluation using RMSE, MAE

Chapter 3: Content-Based Recommendation

• Text preprocessing (tokenization, stopword removal, stemming)
• TF-IDF feature construction for jokes
• Personalized user profiling based on liked items
• Non-personalized item similarity recommendation
• Evaluation using Precision@k, Recall@k and metric visualization

---

Future Improvements

• Model hyperparameter optimization (e.g., dynamic k-tuning)
• Incorporating additional user/item features
• Scaling to larger datasets (e.g., MovieLens)
• Exploring hybrid recommendation approaches

---

License

This project is licensed under the MIT License.
Feel free to fork, adapt, and build upon this work!

---

Acknowledgements

• Course assignments inspiration from RS/ML academic settings.
• NLTK and scikit-learn used for basic text preprocessing and TF-IDF vectorization.