TianyuanYang / KAN4Rec

Implementation of Kolmogorov-Arnold Network (KAN) for Recommendations

Kolmogorov-Arnold Network (KAN) for Recommendations

Official Implementation of Kolmogorov-Arnold Network (KAN) for Recommendations. Any communications, collaborations, issues, PRs are welcomed. The contributors will be listed at contributor part. Please contact [e-mail] or [e-mail]. If you find this repository useful to your research or work, it is really appreciate to star this repository. ❤️

Table of Contents

1. Usage
2. Acknowledgement
3. Citation

Usage

Datasets

MovieLens-1m and MovieLens-20m.

datasets	#users	#items	#actions	average length	density
ML-1m	6040	3416	1m	163.5	4.79%
ML-20m	138,493	26,744	20m	144.4	0.54%

still updating...

Requirements

codes are tested on Python3.8.16 and 1 NVIDIA Tesla V100 SXM2 16 GB

numpy==1.23.5
pandas==1.5.3
scipy==1.9.1
torch==2.0.0
tqdm==4.65.0
wget==3.2

Quick Start

for ml-1m dataset

python main.py --template train_kan4rec --lr 1e-2 --dataset_code ml-1m

for ml-20m dataset

python main.py --template train_kan4rec --lr 1e-2 --dataset_code ml-20m

Results

ML-1m

NDCG	BERT4Rec	KAN4Rec
@1	0.3445	0.3499
@5	0.5068	0.5133
@10	0.5417	0.5477
@20	0.5657	0.5719
@50	0.5875	0.5932
@100	0.5937	0.5991

Recall	BERT4Rec	KAN4Rec
@1	0.3445	0.3499
@5	0.6517	0.6560
@10	0.7590	0.7622
@20	0.8535	0.8575
@50	0.9622	0.9635
@100	0.9997	0.9997

ML-20m

NDCG	BERT4Rec	KAN4Rec
@1	0.5980	0.5982
@5	0.7609	0.7612
@10	0.7796	0.7801
@20	0.7871	0.7871
@50	0.7895	0.7898
@100	0.7906	0.7908

Recall	BERT4Rec	KAN4Rec
@1	0.5980	0.5982
@5	0.8947	0.8949
@10	0.9518	0.9526
@20	0.9799	0.9798
@50	0.9928	0.9928
@100	0.9997	0.9996

still updating...

Acknowledgements

Our code are partly based on the following GitHub repository. Thanks for their awesome works.

• BERT4Rec-VAE-Pytorch: the implement of BERT4Rec model (PyTorch version).
• fast-kan: the implement of KAN (fast version).

Citations

If you find this repository helpful, please cite our paper (coming soon).

Contributors

(back to top)