G-CutMix: A CutMix-based graph data augmentation method for bot detection in social networks.

Area of Science:

• Artificial Intelligence
• Computer Science
• Network Security

Background:

• Data augmentation is crucial for training robust neural networks, particularly in image classification.
• Traditional methods like CutMix are effective for images but not directly applicable to complex graph data.
• Bot detection in social media networks is challenging due to evolving and subtle bot behaviors.

Purpose of the Study:

• To propose G-CutMix, a novel data augmentation method tailored for graph learning.
• To enhance the performance of bot detection systems in social media networks.
• To adapt the CutMix augmentation strategy for graph-structured data.

Main Methods:

• G-CutMix performs CutMix operations between an original graph and a shuffled version.
• It integrates shuffled graph data before the graph convolution process.
• Outputs are merged with user representations from both original and shuffled graphs.

Main Results:

• G-CutMix consistently improves bot detection performance across various Graph Neural Network (GNN) architectures.
• The method enhances performance for Graph Convolutional Networks, GraphSAGE, and Graph Attention Networks.
• The augmentation strategy effectively captures subtle and varied bot behaviors.

Conclusions:

• G-CutMix is an effective data augmentation technique for graph learning in bot detection.
• The approach enhances the robustness and accuracy of GNNs for social media network security.
• G-CutMix offers a promising solution for combating sophisticated bot activities.