README.md

LINE(Large-scale Information Network Embedding)

Introduction

LINE is one of graph embedding methods which preserves both the local and global graph structures. It samples edges from graph to measure first-order proximity and caculates second-order proximity between nodes using their neighbors.

Key points to build a LINE model

• generate traning pairs.
• encode embedding and caculate their similartiy through two LookupEncoders which store two groups of embedding (node embedding and context embedding in the paper)

How to run

1. Prepare data

   cd ../../data/
   python arxiv.py

2. Train

    python train.py

   to train and save embeddings using node2vec.

3. Evaluate

   cd ../../eval/
   python link_prediction_eval.py

Dataset and performance

Dataset	AUC
arxiv	~0.86 (first order)
arxiv	~0.72 (second order) batchsize=32

References

LINE: Large-scale Information Network Embedding