yuanhao53/RG2

A newly designed loss function for collaborative filtering called RG2

Learning-Efficient Yet Generalizable Collaborative Filtering for Item Recommendation (ICML 2024)

The Ranking-Generalizable square Loss (RG^2) method is a newly designed loss function for collaborative filtering that utilizes Alternating Least Squares for optimization.

The experimental results in our work is totally based on the highly-modularized recommendation library RecStudio. To run the code, please follow the instructions:

• Clone the RecStudio repo (https://github.com/ustcml/RecStudio) and install the dependencies, add recstudio to PYTHONPATH.

• Clone this repo and run the run.py