/cs224w-project

Project repo for cs224w

Primary LanguagePythonApache License 2.0Apache-2.0

CS224w

Project repo for Stanford CS224w: Detecting hateful users on Twitter

Preprocess data for pytorch geometric (GNN)

Make sure in the data folder there are users.edges, users_hate_all.content, users_hate_glove.content To create the preprocessed dataset files, create an input folder in data/ then run:

cd code
python data_preprocess_gnn.py --feature glove
python data_preprocess_gnn.py --feature all

In data/input/, there will be two preprocessed dataset files

Run GraphSage/GAT on the retweet graph

cd code
python fs_graphsage.py #runs with all feature vector & graphsage model
python fs_graphsage.py --feature glove #defaults to all features, 320 dimensions
python fs_graphsage.py --model_type gat #defaults to sage, reproduce graphsage experiment

Run LoNGAE on the retweet graph

cd code
git clone https://github.com/vuptran/graph-representation-learning.git
python data_preprocess_LoNGAE.py #get preprocessed data
#Customize model training and metric evaluation
python train_multitask_lpnc.py hateful <gpu_id>

Dataset and original files

Primary dataset: https://www.kaggle.com/manoelribeiro/hateful-users-on-twitter
Dataset owners' analysis repo: https://github.com/manoelhortaribeiro/HatefulUsersTwitter
Their GraphSAGE embedding repo: https://github.com/manoelhortaribeiro/GraphSageHatefulUsers

Their papers:

"Like Sheep Among Wolves":  Characterizing Hateful Users on Twitter
Manoel Horta Ribeiro, Pedro H Calais, Yuri A Santos,  Virgílio AF Almeida, Wagner Meira Jr
MIS2 workshop at WSDM'18

Characterizing and Detecting Hateful Users on Twitter
Manoel Horta Ribeiro, Pedro H Calais, Yuri A Santos,  Virgílio AF Almeida, Wagner Meira Jr
ICWSM'18

GraphMix: Regularized Training of Graph Neural Networks for Semi-Supervised Learning
Verma, V., Qu, M., Lamb, A., Bengio, Y., Kannala, J., & Tang, J. (2019). 

Learning to make predictions on graphs with autoencoders
Tran, P. V. (2018, October). 
In 2018 IEEE 5th International Conference on Data Science and Advanced Analytics (DSAA) (pp. 237-245). IEEE.