DirectSpec

Codes for the follow papers:
Balancing Embedding Spectrum for Recommendation

Environment

• Pytorch+GPU==1.8.0
  
• Numpy==1.19.2
  
• Pandas==1.1.4
  

Run the Algorithm

For GNN encoder, run store_graph_adj.py to generate the grpah embeddings:

 python store_graph_adj.py --dataset=' ' --layer=

• Yelp

 python 'file_name.py' --dataset='yelp' --embedding_size=64 --lr=10 --reg=0.01 --batch_size=256 --alpha=0.8 --tau=3.0 --shrink_norm=0.0

For adaptive temprature desings, set tau_0=2.5, tau_1=3.0

• CiteULike

 python 'file_name.py' --dataset='citeulike' --embedding_size=64 --lr=180 --reg=0.01 --batch_size=256 --alpha=1.0 --tau=3.0 --shrink_norm=0.03

For adaptive temprature desings, set tau_0=2.5, tau_1=3.0

• Gowalla

 python 'file_name.py' --dataset='gowalla' --embedding_size=64 --lr=200 --reg=0.01 --batch_size=256 --alpha=0.7 --tau=4.0 --shrink_norm=0.02

For adaptive temprature desings, set tau_0=3.5, tau_1=4.0