This is the official implementation of Post-Training Embedding Enhancement for Long-Tail Recommendation (CIKM 2024 Short Paper).
The supplementary document provides additional details and experimental results to support the main paper.
We use three datasets, gowalla, yelp2018, and ml10m. Each dataset is split into train/validation/test sets under two test settings, unbiased (fair) and biased (bias) settings. The datasets are in the dataset folder.
python main.py --dataset [fair/bias]_[gowalla/yelp2018/ml10m]
For example, to run gowalla on the ubiased test setting, run:
python main.py --dataset fair_gowalla
The learned embeddings are in embs folder.
python edge.py --dataset [fair/bias]_[gowalla/yelp2018/ml10m] --alpha [alpha] --beta [beta] --tau [tau] --lmbda [lambda]
For example, to run ml10m on the biased test setting, run:
python edge.py --dataset bias_ml10m --alpha 0.8 --beta 0.2 --tau 0.2 --lmbda 0.4
The configuration of each dataset can be found in run.sh. You can simply execute ./run.sh.
This code is implemented based on the open source code from the paper Test Time Embedding Normalization for Popularity Bias Mitigation.