Giters

sisinflab / MORS_reproducibility

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Reproducibility of Multi-Objective Reinforcement Learning Recommendation: Interplay between Effectiveness and Beyond-Accuracy Perspectives

This repository refers to source codes and datasets of the reproducibility study Reproducibility of Multi-Objective Reinforcement Learning Recommendation: Interplay between Effectiveness and Beyond-Accuracy Perspectives accepted as Reproducibility Paper at The 17th ACM Recommender Systems Conference (RecSys 2023).

Due to the size limits of GitHub, we load the files at this link. This link contains two folders:

  1. SMORL: it contains the source codes and datasets to reproduce the results in Section 4;
  2. my_sir_elliot: it contains the source codes and datasets to reproduce the results in Section 5.

Please, make sure to have a Python 3.8.0 or later version installed on your device.

We advice to treat these folders as two separate projects. For each project, you may create the virtual environment with the requirements files we included as follows:

$ python3 -m venv venv
$ source venv/bin/activate
$ pip install --upgrade pip
$ pip install -r requirements.txt

To run the models in the SMORL project, you should install a version of pytorch compatible with the CUDA version of your device.

SMORL Project

In the SMORL Project, you find two folders:

  1. div4rec contains datasets and files needed to run the models and the results gathered in Section 4;
  2. src contains the source codes to run the models.

To run a model, you may run the following commands:

$ cd src
$ python -u {modelname}.py

where {modelname} is one of the following: Caser, CaserSMORL, GRU, GRUSMORL, NexItNet, NextItNetSMORL, SASRec, SASRecSMORL.

In these files, you should change the datasets name in the Args class. Please, you should use rc15 for the RC 15 dataset, while retail_rocket for the Retailrocket dataset. You will find the results in the corresponding folder within the div4rec directory.

my_sir_elliot Project

This project contains an ad-hoc version of the Elliot framework. For more details, please refer to its official documentation.

With this project, you reproduce the performance in terms of Bias Disparity, PopRSP, and PopREO. To do this, you may run the following command:

$ python -u start_experiments.py

In this file, you should select one line between lines 9 and 10 at your convenience to obtain the results on RC 15 and Retailrocket datasets, respectively. You will find the results in results/{dataset}/performance folder.

About