BACO

Balanced Co-Clustering of Users and Items for Embedding Table Compression in Recommender Systems

Prerequisites

---

• Please refer to requirements.txt to install all the dependencies required for running the algorithm
• Please enter the code directory before running the command.
• Please run \code\setup.sh to compile the core components implemented in Cython.

Dataset

• Preprocessed datasets are available at \code\datasets

• Raw datasets: $Beauty$ is from the paper $\texttt{DirectAU}$ (link); $Gowalla$, $Yelp2018$, and $AmazonBook$ are from the paper $\texttt{GraphHash}$ (link).

Dataset detail

---

We conduct experiments on four datasets:

• $Beauty$: A dataset presenting user feedback and ratings for beauty products.
• $Gowalla$: A social network dataset with user check-in data.
• $Yelp2018$: A dataset containing user reviews and ratings for businesses.
• $AmazonBook$: A dataset containing user reviews and ratings for books.

Run experiments

---

Example command to run $\texttt{BACO}$:

cd code
python train.py model=LightGCN hash_type=BACO dataset=Gowalla resolution=7.57

More commands for running are provided in \code\commands.sh, with the best searched hyperparameters.