This project was developed by Cesare Bernardis, Ph.D. candidate at Politecnico di Milano, and it is based on the recommendation framework used by our research group. The code allows reproducing the results of "Eigenvalue Perturbation for Item-based Recommender Systems", a work presented at the ACM RecSys Conference 2021.
Please cite our article (BibTex, ACM DL) if you use this repository or a part of it in your work.
To run the experiments, we suggest creating an ad hoc virtual environment using Anaconda.
You can also use other types of virtual environments (e.g. virtualenv). Just ensure to install the packages listed in 'requirements.txt' in the environment used to run the experiments.
To install Anaconda you can follow the instructions on the website.
To create the Anaconda environment with all the required python packages, run the following command:
bash create_env.shThe default name for the environment is recsys-eigenpert. You can change it by replacing the string recsys-eigenpert in 'create_env.sh' with the name you prefer. All the following commands have to be modified according to the new name.
Note that this framework includes some Cython implementations that need to be compiled. Before continuing with next steps, ensure to have a C/C++ compiler installed. On several Linux distributions (e.g. Ubuntu, Debian) it is enough to run the following:
sudo apt-get install build-essentialFinally, to install the framework you can simply activate the created virtual environment and execute the 'install.sh' script, using the following commands:
conda activate recsys-eigenpert
sh install.shFirst of all, you should activate the environment where you installed the framework. If you followed the installation guide above, the command is:
conda activate recsys-eigenpertThen, the first step to run the experiments consists in downloading the data and generating the splits. These actions are performed by the 'create_splits.py' script that can be executed with the following command:
python create_splits.pyTo perform the evaluation on the splits created, run:
python scripts/run_test.pyThe results are saved in a CSV file called 'optimized_results.csv'.
Automatically, the script employs the best hyper-parameter configurations found by the optimization procedure we performed in our experiments. You can find these configurations in 'data/best-hyperparameters.zip'.
If you want to perform a new hyper-parameter optimization, you can use:
python scripts/run_parameter_search.py