Personalized Recommendation System with Matrix Factorization and Bias Terms

This repository contains the code and resources for building a personalized movie recommendation system using matrix factorization, incorporating both latent factors and bias terms. The model is applied to the MovieLens 25M dataset and evaluated on different hyperparameters and configurations.

Project Structure

├── assets/                   # Generated images (e.g.visualizations )
├── main.ipynb                # Main notebook for project work on the 25M dataset
├── ml-25M/                   # MovieLens 25M dataset (excluded from the repo due to size)
├── ml-latest-small/          # Small dataset used for prototyping (excluded from the repo due to size)
├── models/                   # Model files (pickled models, excluded from the repo due to size)
├── prototype.ipynb           # Prototype code used for quick testing on small dataset
├── README.md                 # This file
├── requirements.txt          # Python dependencies

Overview

This project explores matrix factorization techniques to build a movie recommendation system. The approach includes:

• Matrix Factorization: Used to capture latent factors for users and items.
• Bias Terms: Incorporated for both users and items to improve model accuracy.
• Evaluation: The model is tested using RMSE and the results are compared across different embedding dimensions (k).

The full MovieLens 25M dataset was used for the final model, while a smaller dataset (ml-latest-small) was used for prototyping.

Setup

1. Clone this repository:

   git clone https://github.com/Bwhiz/Recommender-System-ALS.git
   cd Recommender-System-ALS

2. Install the required dependencies:

   pip install -r requirements.txt

3. Download the datasets:

   The datasets (ml-25M and ml-latest-small) are too large to be included in the repository. You can download the 25M dataset from here, while you can download the small dataset here. Place them in the respective folders.

4. The pickle files would be generated when the main.ipynb file is executed and it would be saved in the 'models/' folder.

Running the Notebooks

• Prototype.ipynb: This notebook contains the code for prototyping the recommendation system on the smaller dataset (ml-latest-small).

• Main.ipynb: This notebook contains the complete workflow, including data preprocessing, model building, and evaluation on the full MovieLens 25M dataset.

Results

The results from the final model, including the effects of different embedding dimensions (k), can be found in the generated images in the assets/ folder.