NRL-implement

Re-implementation of four Network representation learning (NRL) algorithms: DeepWalk, LINE, node2vec, GraphGAN.

Environment

• NumPy
• TensorFlow
• gensim
• NetworkX

Data

There are two datasets located in the path ./data/:

• Cora: citation dataset.
• Tencent Weibo: following network.

Training

First, locate at the root path of the project:

cd NRL-implement

For DeepWalk:

python DeepWalk/main.py

For LINE:

python LINE/main.py

For node2vec:

python node2vec/main.py

These three implementations use cora as dataset, and results are saved in ./results/cora/.

Use logistic regression as classifier to evaluate the quality of embeddings produced by these three implementations.

python LRclassifier.py --method DeepWalk

where DeepWalk can be replaced by LINE and node2vec.

For GraphGAN:

python GraphGAN/main.py

which uses tencent as dataset.