AttRec: A Self-Attention Recommender System

This repository implements a self-attention-based recommender system, AttRec, for next-item recommendation. The project includes model implementation, evaluation, and tuning scripts, alongside Jupyter notebooks for detailed experimentation.

Link: https://arxiv.org/pdf/1808.06414

---

Table of Contents

1. Project Overview
2. Project Structure
3. Setup and Installation
4. Experimentation
5. References
6. Acknowledgments

---

Project Overview

AttRec uses a self-attention mechanism to model sequential user behavior for accurate next-item predictions. It is designed for flexibility, scalability, and ease of experimentation.

Key features:

• End-to-end pipeline: Data preprocessing, model training, and evaluation scripts.
• Pretrained models: Included checkpoint files for quick testing.
• Jupyter notebooks: For visual exploration and detailed analyses.

---

Project Structure

.
├── attrec.ipynb              # Main notebook demonstrating AttRec implementation
├── ncf_deep_dive.ipynb       # Notebook exploring Neural Collaborative Filtering
├── processed_data/           # Preprocessed training and test datasets
│   ├── info.pkl
│   ├── test.csv
│   └── train.csv
├── pyproject.toml            # Python project configuration
├── recommenders/             # Core implementation directory (From https://github.com/recommenders-team/recommenders)
│   ├── README.md             # Detailed documentation for this module
│   ├── datasets/             # Data loading and processing scripts
│   ├── evaluation/           # Evaluation metrics and scripts
│   ├── models/               # Model architecture definitions
│   ├── tuning/               # Hyperparameter tuning scripts
│   └── utils/                # Utility functions
├── requirements.txt          # Dependencies for the project
├── save_path/                # Directory for saving model checkpoints
└── setup.py                  # Setup script for the project

---

Setup and Installation

Prerequisites

• Python 3.8 or higher
• GPU-enabled machine (recommended for training)

Installation

For environment management, we recommend using conda, and for development, VS Code is suggested. Follow these steps to install the recommenders package and run a sample notebook on Linux/WSL:

1. Install GCC
   If GCC is not already installed, you can install it on Ubuntu with:

   sudo apt install gcc

2. Set Up a Conda Environment
   Create and activate a new Conda environment:

   conda create -n <environment_name> python=3.9
   conda activate <environment_name>

3. Install the Recommenders Package
   Install the core recommenders package to run all CPU-compatible notebooks:

   pip install recommenders

4. Create a Jupyter Kernel
   Set up a Jupyter kernel for the environment:

   python -m ipykernel install --user --name <environment_name> --display-name <kernel_name>

5. Clone the Repository
   Clone the repository using VS Code or the command line:

   git clone https://github.com/recommenders-team/recommenders.git

6. Run an Example Notebook in VS Code

   • Open a notebook, such as attrec.ipynb.
   • Select the Jupyter kernel <kernel_name>.
   • Run the notebook.

This setup ensures you have everything needed to start working with the recommenders package efficiently.

---

Experimentation

Use the included notebooks for interactive exploration:

• attrec.ipynb: Full implementation walkthrough for AttRec.
• ncf_deep_dive.ipynb: Insights into Neural Collaborative Filtering. (From https://github.com/recommenders-team/recommenders)

---

References

• Shuai Zhang et al., "Next Item Recommendation with Self-Attention."
• S. Ge, "AttRec: A Recommender System with Self-Attention Mechanism," slientGe/AttRec

---

Acknowledgments

• This project is inspired by and adapted from the AttRec implementation by S. Ge.
• Special thanks to the authors of Next Item Recommendation with Self-Attention for the foundational research.
• Special thanks to the recommenders-team for the recommender template recommenders-team/recommenders
• This project is part of the 2301491 SPECIAL TOPICS IN COMPUTER SCIENCE (Popular Techniques in Recommender Systems) course at Chulalongkorn University in 2024.

---