Social Bot Identify Method Based on Deep Graph Convolutional Network

This paper presents a method of social robot recognition. This method extracts the characteristics of account sentiment diversity and uses the RoBERTa (robustly optimized BERT pretraining approach) model to classify the sentiment of blog posts. At the same time, the single-pass method is used to cluster blog posts and construct blog similarity graph. On this basis, attention-GCNII (A-GCNII) model, which adds Attention mechanism on the basis of graph convolutional network via initial residual and identity mapping (GCNII) model, is proposed to identify social robots by capturing user metadata features and user relationship structure features under the same topic in social networks. The results of comparative experiments on real Sina Weibo datasets show that our proposed method performs well in recognition accuracy and recognition effect.