Jerry-zhxf / Classification-of-Social-Network-Node-based-on-Graph-Convolutional-Network

Classification of Social Network Node based on Graph Convolutional Network

GCN/GraphSAGE_Project - comp7404-group4

Team Member: Wang Luozhou, Zhang He, Ji Yeon Fung, Ren Xinzhu

Language Environment:

Python3.6.9

Setup

pip install -r requirements.txt 

PS: If you get an error report in the dgl_cu101 ==0.5.2 command, please delete it and run the following command:

pip install --verbose --no-cache-dir torch-sparse torch-scatter

Because you probably don't have a graphics card or version of the driver, you cannot use CUDA to train your model. Therefore, you'll need to install the environment packages(Torch-Sparse and Scatter) to make sure that the code works (This model also can be trained directly with the CPU).

Run Demo

Run three datasets by type command in terminal

1. Cora Dataset

python run.py --dataset Cora 

2. CiteSeer Dataset

python run.py --dataset CiteSeer 

3. PubMed Dataset

python run.py --dataset PubMed 

The output will plot a curve of accuracy and loss firstly. When user switches off this window, the output will be two figures showing nodes position through t-SNE visualization before input into model and after respectively. Here are samples.

If user want to use different of aggregators of GraphSAGE, use command

1. gcn aggregator

python run_aggregator.py --aggregator gcn 

2. MaxPooling aggregator

python run_aggregator.py --aggregator pool 

3. LSTM aggregator

python run_aggregator.py --aggregator lstm 

User can also add --dataset to select different dataset among Cora, CiteSeer, PubMed.