HERECJ / AdaSIR

Implementation for AdaSIR

"Learning Recommenders for Implicit Feedback with Importance Resampling" (WWW2022) The codes are tested in Pytorch

Parameters

• data
  • gowalla, yelp, amazoni
• d
  • embedding size
• m, model
  • 0: matrix factorization
  • 1: NCF
  • 2: GMF
  • 3: MLP
• sampler
  • 0: uniform
  • 2: AdaSIR uniform
  • 3: popularity
  • 5: AdaSIR pop
  • 7: AdaSIR uniform + rank estimation
  • 8: DNS
  • 9: Adaptive kernel(only works for matrix factorization)

Running Example

python main_more.py --sampler 0 --weighted for PRIS(U)

python main_more.py --sampler 2 --weighted for AdaSIR

or

sh run.sh