LinUCRL for Recommender Systems

Based on paper "Fighting Boredom in Recommender Systems with Linear Reinforcement Learning", Romain WARLOP, Alessandro Lazaric, Jérémie Mary.

This repository contains:

• an implementation of LinUCRL algorithm;
• preprocessing functions for running on MovieLens-1m dataset.

Requirements

Requires Python 3.6 and tested on Ubuntu 16.04. Please check out requirements.txt for resolving dependency issues.

Quick start

1. Clone this repository
2. Download MovieLens dataset

   make -B download_dataset
   

3. Run an experiment

   make -B train
   

Configuration

Check out configuration file lucrl/config/config.yaml which contains parameters for dataset, mdp and LinUCRL algorithm.

Acknowledgments

1. All the experiments was initially run using Ocean framework for Data Science research.
2. Thanks to Romain for providing me with original implementation when I got stuck.