Why is Normalization Necessary for Linear Recommenders? (SIGIR'25)

This is the official code for DAN in the paper "Why is Normalization Necessary for Linear Recommenders?", The 48th International ACM SIGIR Conference on Research and Development in Information Retrieval.

Introduction

Despite their simplicity, linear autoencoder (LAE)-based models have shown comparable or even better performance with faster inference speed than neural recommender models. However, LAEs face two critical challenges:

1. Popularity bias: LAEs tend to recommend popular items excessively
2. Neighborhood bias: LAEs overly focus on capturing local item correlations

To address these issues, we propose Data-Adaptive Normalization (DAN), a versatile normalization solution that flexibly controls the popularity and neighborhood biases by adjusting item- and user-side normalization to align with unique dataset characteristics.

Note that a summary of our paper is on our lab blog (in Korean).

Requirements

pip install -r requirements.txt

How to run

• Run LAE with DAN

sh dan_lae.sh

• Run EASE with DAN

sh dan_ease.sh

• Run RLAE with DAN

sh dan_rlae.sh

TODO

• Add the ML-20M, Netflix, and MSD datasets
• Add weak generalization protocol

Citation

If you find our work useful for your research, please cite our paper:

@inproceedings{park2025dan,
  title={Why is Normalization Necessary for Linear Recommenders?},
  author={Seongmin Park and
          Mincheol Yoon and
          Hye-young Kim and
          Jongwuk Lee},
  booktitle={Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval},
  year={2025}
}