RoSGAS : Adaptive Social Bot Detection with Reinforced Self-supervised GNN Architecture Search

Abstract

Social bots are referred to as the automated accounts on social networks that make attempts to behave like human. While Graph Neural Networks (GNNs) has been massively applied to the field of social bot detection, a huge amount of domain expertise and prior knowledge is heavily engaged in the state-of-the art approaches to design a dedicated neural network architecture for a specific classification task. Involving oversized nodes and network layers in the model design, however, usually causes the over-smoothing problem and the lack of embedding discrimination. In this paper, we propose RoSGAS , a novel R einf o rced and S elf-supervised G NN A rchitecture S earch framework to adaptively pinpoint the most suitable multi-hop neighborhood and the number of layers in the GNN architecture. More specifically, we consider the social bot detection problem as a user-centric subgraph embedding and classification task. We exploit heterogeneous information network to present the user connectivity by leveraging account metadata, relationships, behavioral features and content features. RoSGAS uses a multi-agent deep reinforcement learning (RL) mechanism for navigating the search of optimal neighborhood and network layers to learn individually the subgraph embedding for each target user. A nearest neighbor mechanism is developed for accelerating the RL training process, and RoSGAS can learn more discriminative subgraph embedding with the aid of self-supervised learning. Experiments on 5 Twitter datasets show that RoSGAS outperforms the state-of-the-art approaches in terms of accuracy, training efficiency and stability, and has better generalization when handling unseen samples.

SEGCN: a subgraph encoding based graph convolutional network model for social bot detection

“…Recently, RoSGAS 47 designed an adaptive search GNN structure for social bot detection model, which gets rid of the a priori of people designing GNN structures and searches for appropriate GNN structures through reinforcement learning. RF-GNN 48 utilized the idea of integrated learning to detect social bots by combining the Random Forest algorithm and GNN.…”

Multimodal detection of epilepsy with deep neural networks

“…A reinforcement learning approach was proposed by [35] for searching the GNN architecture. In this way, the most suitable multi-hop neighborhood and the number of layers in the GNN architecture are found.…”

FedACK: Federated Adversarial Contrastive Knowledge Distillation for Cross-Lingual and Cross-Model Social Bot Detection

“…The main task is to let user-level backbone feature extractor model 𝜀 to extract features per user metadata (e.g., account properties) and textual data (e.g., tweets). We concatenate the key items extracted from the metadata into the property vector 𝑢 𝑝 , following the similar way as [41,42], which is converted into user's property representation 𝑟 𝑝 by a Multi-layer Perceptron 𝑟 𝑝 = 𝑀𝐿𝑃 (𝑢 𝑝 ). For textual data, assume a user has posted M Tweets…”

“…There is a new yet understudied problem in bot detection -a society of bots tend to be exposed to multiple social platforms and behave as collaborative cohorts. Existing bot detection solutions largely rely on user property features extracted from metadata [9,41], or features derived from textual data such as a tweet post [15,39], before adopting graph-based techniques to explore neighborhood information [14,42,46]. While such models can uncover camouflage behaviors, they are siloed and subject to the amount, shape, and quality of platform-specific data.…”