/NASR

Pytorch implementation of NASR

Primary LanguagePython

(WWW'22) Towards Automatic Discovering of Deep Hybrid Network Architecture for Sequential Recommendation

🙋 Please let us know if you find out a mistake or have any suggestions!

🌟 If you find this resource helpful, please consider to star this repository and cite our research:

@inproceedings{cheng2022towards,
  title={Towards automatic discovering of deep hybrid network architecture for sequential recommendation},
  author={Cheng, Mingyue and Liu, Zhiding and Liu, Qi and Ge, Shenyang and Chen, Enhong},
  booktitle={Proceedings of the ACM Web Conference 2022},
  pages={1923--1932},
  year={2022}
}

Introduction

In this work, we use network architecture search(NAS) with self-supervised learning to find a hybrid&suitable architecture on a given dataset adaptively in sequential recommendation. The framework is illustrated as follows:

model

Requirements

  • pytorch 1.x
  • pandas
  • numpy
  • tqdm

Usage

python main.py --data_path your_path --max_len seq_len

For more detailed params, please refer to args.py

Results

We performed our NASR on two datasets of Movielens and LastFM, and the result tested on Movielens is shown as follows:

for fair comparison, all the models are trained in auto-regressive manner

NDCG@20 Recall@20 NDCG@10 Recall@10
GRU4Rec 0.1587 0.3344 0.1347 0.2390
Nextitnet 0.1653 0.3421 0.1413 0.2471
SASRec 0.1665 0.3428 0.1428 0.2485
Ramdom search 0.1605 0.3395 0.1361 0.2426
NASR 0.1793 0.3468 0.1548 0.2676

Citation

@inproceedings{cheng2022towards,
  title={Towards automatic discovering of deep hybrid network architecture for sequential recommendation},
  author={Cheng, Mingyue and Liu, Zhiding and Liu, Qi and Ge, Shenyang and Chen, Enhong},
  booktitle={Proceedings of the ACM Web Conference 2022},
  pages={1923--1932},
  year={2022}
}