Social bots detection model based on joint sampling and jumping knowledge networks

The widespread existence of social bots in social networking platforms seriously threatens network security. Graph-based detection models significantly enhance the detection of social bots by incorporating both the topological structure and user characteristics of social networks. Considering the limitations of existing graph neural network models in sampling strategies and the significant correlation between node features and graph structures, this paper introduces a social bots detection model that integrates joint sampling techniques and jumping knowledge networks. The model initially employs a composite sampling unit to acquire the sequence of neighboring nodes most relevant to the target node. Subsequently, it utilizes this sequence in conjunction with a jumping knowledge graph convolution to update the node features. Finally, a jumping knowledge networks model assigns appropriate neighborhood information to each node to complete the classification task. The accuracy of proposed model achieves 92.92% on Weibo dataset, which is superior to benchmark methods.