GammaGL/examples/tadw

• Network Representation Learning with Rich Text Information (TADW)
• Dataset Statics
• Performance

Network Representation Learning with Rich Text Information (TADW)

• Paper link: https://www.ijcai.org/Proceedings/15/Papers/299.pdf
• Author's code repo: https://github.com/albertyang33/TADW. Note that the original code is
  implemented with MATLAB for the paper.

Dataset Statics

Dataset	# Nodes	# Edges	# Classes
Cora	2,708	10,556	7
Citeseer	3,327	9,228	6

Refer to Planetoid.

Performance

For all the datasets: The training ratio is 50% for linear SVM.

Dataset	Paper(10%)	Paper(20%)	Paper(30%)	Paper(40%)	Paper(50%)	Our(tf)	Our(th)	Our(pd)	Our(ms)
Cora	82.4	85.0	85.6	86.0	86.7	84.43(±0.37)	84.42(±0.96)	84.34(±0.64)	84.08(±0.53)
Citeseer	70.6	71.9	73.3	73.7	74.2	74.09(±1.11)	74.41(±0.58)	73.92(±0.72)	73.87(±0.57)

TL_BACKEND="tensorflow" python tadw_trainer.py --dataset cora --lr 0.2 --n_epoch 100 --embedding_dim 80 --lamda 0.2 --svdft 200 
TL_BACKEND="torch" python tadw_trainer.py --dataset cora --lr 0.2 --n_epoch 100 --embedding_dim 80 --lamda 0.2 --svdft 200
TL_BACKEND="paddle" python tadw_trainer.py --dataset cora --lr 0.2 --n_epoch 100 --embedding_dim 80 --lamda 0.2 --svdft 200
TL_BACKEND="mindspore" python tadw_trainer.py --dataset cora --lr 0.2 --n_epoch 100 --embedding_dim 80 --lamda 0.2 --svdft 200
TL_BACKEND="torch" python tadw_trainer.py --dataset citeseer --lr 0.1 --n_epoch 50 --embedding_dim 500 --lamda 0.5 --svdft 300
TL_BACKEND="tensorflow" python tadw_trainer.py --dataset citeseer --lr 0.1 --n_epoch 50 --embedding_dim 500 --lamda 0.5 --svdft 300
TL_BACKEND="paddle" python tadw_trainer.py --dataset citeseer --lr 0.1 --n_epoch 50 --embedding_dim 500 --lamda 0.5 --svdft 300
TL_BACKEND="mindspore" python tadw_trainer.py --dataset citeseer --lr 0.1 --n_epoch 50 --embedding_dim 500 --lamda 0.5 --svdft 300