caboose

This repository contains the code for our recent SIGIR paper on unlearning data from kNN recommendations models:

Sebastian Schelter, Mozhdeh Ariannezhad, Maarten de Rijke (2023). Forget Me Now - Fast and Exact Unlearning in Neighborhood-Based Recommendation. ACM SIGIR'23

This code contains the Python frontend and the integration into next-basket recommendation algorithms. The Rust code for the underlying index is available at https://github.com/schelterlabs/caboose_index .

Installation for Development

Preliminaries

• Make sure you have Python 3.9 and Rust installed: https://www.rust-lang.org/tools/install
• Make sure to have https://github.com/schelterlabs/caboose_index checked out as caboose_index in the same folder as this project

Dependencies and build

• Setup a virtualenv python3.9 -m venv venv and source venv/bin/activate
• Install Cython (needed for similaripy) pip install Cython==0.29.32
• Install the dependencies pip install -r requirements.txt
• Build the project with maturin develop --release

Validation

• Run the tests with pytest
• Start jupyter and run the minimal notebook