DeepWalk samples sequences of nodes from the given graph, and use skip-grapm model to train node embeddings.
- sample sequences and generate pairs.
- use LookupEncoders to encode EgoGraphs to node embedding.
- Prepare data
cd ../../data/ python blogcatelog.py
- Train
python train.py
- Evaluate
to train classfier and get F1 score.
cd ../../eval/ python blogcatelog_eval.py
Dataset | macro F1 |
---|---|
BlogCatalog | ~0.23 (50% labeled nodes) |