Implementation for DNS(M,N) and Softmax(\rho, N)

This is our PyTorch implementation for the paper 2023'WWW:

Wentao Shi, Jiawei Chen, Fuli Feng, Jizhi Zhang, Junkang Wu, Chongming Gao, Xiangnan He (2023) On the Theories behind Hard Negative Sampling for Recommendation. paper link. In WWW 2023.

Dependencies

• Compatible with PyTorch 1.8.2 and Python 3.8.
• Dependencies can be installed using requirements.txt.

Parameters

• data: The dataset to use, which can be:
  • gowalla
  • yelp
  • amazoni
• d: The embedding size
• m: The model to use, which can be:
  • 0: matrix factorization
  • 1: NCF
  • 2: GMF
  • 3: MLP
  • 4: LightGCN
• sampler: The sampling method to use, which can be:
  • 0: uniform
  • 2: AdaSIR uniform
  • 3: popularity
  • 5: AdaSIR pop
• loss_type: The loss function to use, which can be:
  • 0: AdaSIR
  • 1: DNS(M, N)
  • 2: Softmax(\rho, N)

Running Example

python main_more.py --lambda_w 2 --sampler 0 --sample_num 200 --fix_seed --weighted --loss_type 1 for DNS($M$, $N$)

python main_more.py --lambda_w 1 --sampler 0 --sample_num 200 --fix_seed --weighted --loss_type 2 for Softmax($\rho$, $N$)

Acknowledgement

The project is built upon AdaSIR

For any clarification, comments, or suggestions please create an issue or contact me.