Welcome to the official code repository for the ICWSM 2025 paper "Scaled Supervision is an Implicit Lipschitz Regularizer". It adapts from the public code repository RecBole.

In modern social media, recommender systems (RecSys) rely on the click-through rate (CTR) as the standard metric to evaluate user engagement. CTR prediction is traditionally framed as a binary classification task to predict whether a user will interact with a given item. However, this approach overlooks the complexity of real-world social modeling, where user, item, and their interactive features change dynamically in fast-paced online environments. This dynamic nature often leads to model instability, reflected in overfitting short-term fluctuations rather than higher-level interactive patterns. In this work, we aim to alleviate the overfitting problem by increasing the supervision bandwidth in CTR training. We (i) theoretically, we formulate the impact of fine-grained preferences on model stability as a Lipschitz constrain; (ii) empirically, we discover that scaling the supervision bandwidth can act as an \textit{implicit} Lipschitz regularizer, stably optimizing existing CTR models to achieve better generalizability.

1. Install all the required packages found here.
2. Download all the benchmark datasets used in the paper, including ML-1M, Yelp2018, and Amazon-book. Make sure the folders are named ml-1m, yelp2018, and amazon-book respectively.
3. Run the command python run_recbole.py --model=<model_name> --dataset=<dataset_name> [hyper_parameter_setting] to train a new CTR model from scratch.
4. To apply our approach, add the hyperparameter settings --multi_cls=5 --multi_ratio=<chosen_ratio>. You may consult the commands in commands.sh.
5. To evaluate the ranking metrics, python recbole_rank_eval.py --checkpoint=<saved_model_checkpoint>.

@inproceedings{ouyang2025scaled, 
    title={Scaled Supervision is an Implicit Lipschitz Regularizer}, 
    author={Ouyang, Zhongyu and Zhang, Chunhui and Jia, Yaning and Vosoughi, Soroush}, 
    booktitle={Proceedings of the International AAAI Conference on Web and Social Media (ICWSM)}, 
    year={2025}
}

This repository adapts from the public code repository RecBole.