DeepWalk

Introduction

DeepWalk samples sequences of nodes from the given graph, and use skip-grapm model to train node embeddings.

Key points to build a DeepWalk

sample sequences and generate pairs.
use LookupEncoders to encode EgoGraphs to node embedding.

How to run

Prepare data
```
cd ../../data/
python blogcatelog.py
```
Train
```
python train.py
```
Evaluate
```
cd ../../eval/
python blogcatelog_eval.py
```
to train classfier and get F1 score.

Dataset and preformance

Dataset	macro F1
BlogCatalog	~0.23 (50% labeled nodes)

References

DeepWalk: Online Learning of Social Representations