cycl2018 / recommendation-editing

Recommendation Editing

The goal of recommendation editing is to quickly correct the erroneous recommendations of the recommendation system, thereby improving the user-friendliness of the recommendation system. For example, in the real-time re-ranking stage of industrial-level multi-stage recommendation systems, the recommendation system needs to perform real-time editing based on the negative feedback provided by users in real time, reducing the occurrence of negative feedback behavior in the recommended results of the next refresh.

🚀Quick Start

Clone

git clone [email protected]:cycl2018/Recommendation-Editing.git

Set up the required environment

numba==0.58.1
numpy==1.26.3
scipy==1.12.0
torch==2.1.0+cu121

Train the original recommendation model (if you directly use the checkpoint we provide, you can skip it)

• Example of training XSimGCL model by KuaiRand dataset.
• You can refer to the files in the/conf folder for configuration

python train.py --conf conf/XSimGCL/KuaiRand.conf

Editing

• Example of editing XSimGCL model by FT method.
• --best_param indicates running with optimal parameters

python edit.py --model_conf conf/XSimGCL/KuaiRand.conf --edit_type FT --best_param --edit_num 10

Acknowledgments

We are grateful to the authors of SELFRec for making their project codes publicly available.

Citation

Our paper on this benchmark will be released soon!