CREU

This repository is the official implementation of the paper: Harnessing Light for Cold-Start Recommendations: Leveraging Epistemic Uncertainty to Enhance Performance in User-Item Interactions from CIKM 2025.

Model

Requirements

The code is tested under the following environment:

- Python 3.12
- PyTorch 2.7.1
- numpy 2.2.6
- pandas 2.3.0
- scipy 1.17.0
- scikit-learn 1.7.0
- geomloss 0.2.6
- pykeops 2.3

Create a conda environment:

conda create -n creu python=3.12
conda activate creu

# Default PyTorch version is CPU version, you can change it to GPU version according to your needs
# Comment out the CPU version and uncomment the GPU version in requirements.txt
conda install --yes --file requirements.txt

Dataset

We tested our model on two datasets: MovieLens-1M and Taobao Display Ad Click.

Data Statistics

Dataset	#Users	#Items	#Interactions	Sparsity
MovieLens-1M	6,040	3,706	1,000,209	95.8100%
TaobaoAD	1,140,000	850,000	26,550,000	99.9973%

Preprocessing

The raw data needs to be preprocessed using movielens1M_preprocess.ipynb and taobaoAD_preprocess.ipynb in the datahub folder.

You can download processed data from our OneDrive [ MovieLens-1M | TaobaoAD ]

Usage

python main.py --dataset_name <dataset_name>  --model_name <backbone> --warmup_model <model_name> --creu_iters 10

Citation

@inproceedings{CREU,
  author = {Xiang, Yang and Fan, Li and Yin, Chenke and Kong, Menglin and Ji, Chengtao},
  title = {Harnessing Light for Cold-Start Recommendations: Leveraging Epistemic Uncertainty to Enhance Performance in User-Item Interactions},
  year = {2025},
  isbn = {9798400720406},
  publisher = {Association for Computing Machinery},
  doi = {10.1145/3746252.3760793},
  pages = {5361–5365},
  numpages = {5},
  keywords = {cold-start recommendation, recommender system, uncertainty quantification},
  location = {Seoul, Republic of Korea},
  series = {CIKM '25}
}