RGT

Heterogeneity-aware Twitter Bot Detection with Relational Graph Transformers

---

• authors: Shangbin Feng, Zhaoxuan Tan, Rui Li, Minnan Luo

• link: https://arxiv.org/abs/2109.02927

• file structure:

├── cresci-15
│   ├── Heterobot.py  # train model on cresci-2015
|   ├── Dataset.py
|   └── layer.py
├── Twibot-20    
│   ├── Heterobot.py  # train model on Twibot-20
|   ├── Dataset.py
|   └── layer.py
└── Twibot-22
    ├── Heterobot_sample.py  # train model on Twibot-22
    └── layer.py

• Requirements

CUDA==10.2
pytorch-lightning==1.4.9
torch==1.9.1
torch-geometric==2.0.2
torch-cluster==1.5.9
torch-scatter==2.0.9
torch-sparse==0.6.12
torch-spline-conv==1.2.1
sklearn
argparse

• implement details:

1. The input features of our implementation is the same as BotRGCN, for more details about preprocessing please refer to BotRGCN.

2. To enable the training on large-scale graph such as Twibot-22, we adopt the NeighborLoader in torch geometric with num_neighbors set to 20 and sample for 2 iterations.

How to reproduce:

1. run the preprocess code to get following files

cat_properties_tensor.pt
des_tensor.pt
edge_index.pt
edge_type.pt
label.pt
num_properties_tensor.pt
test_idx.pt
train_idx.pt
val_idx.pt
test_idx.pt

2. train the model

Please change the path to proprocess file in the following commands

• on Twibot-20 & cresci-2015

CUDA_VISIBLE_DEVICES=0 python Heterobot.py --batch_size 128 --epochs 200 --path /path/to/preprocessed/file

• on Twibot-22

CUDA_VISIBLE_DEVICES=0 python Heterobot_sample.py --batch_size 256 --epochs 200 --path /path/to/preprocessed/file

Results:

dataset		acc	precison	recall	f1
Cresci-2015	mean	0.9715	0.9638	0.9923	0.9778
Cresci-2015	std	0.0032	0.0059	0.0015	0.0024
Twibot-20	mean	0.8657	0.8515	0.9106	0.8801
Twibot-20	std	0.0041	0.0028	0.0080	0.0041
Twibot-22	mean	0.7647	0.7503	0.3010	0.4294
Twibot-22	std	0.0045	0.0085	0.0017	0.0049

baseline	acc on Twibot-22	f1 on Twibot-22	type	tags
RGT	0.7647	0.4294	F T G	Graph Neural Networks