Same author detection problem
- python (>=3.6)
- pytorch (>=1.4)
- pytorch_geometric (https://github.com/pyg-team/pytorch_geometric)
conda env create --file environment.yaml
- initial paper/author node embedding
- nn.Embedding whose input dimension equals to the total number of paper/authors
- node embedding extraction
- GCN
- link prediction
- minimizing triplet loss (positive pairs gets closer, negative pairs gets distant)
- a bilinear projection to measure the distance of two embeddings
python train_baseline.py