Detecting Social Media Bots with Variational AutoEncoder and k-Nearest Neighbor

Malicious social media bots are disseminators of malicious information on networks and seriously affect security the network environment. Efficient reliable classification is crucial for detecting manipulation in networks. Aiming to correct defects high-cost labeling unbalanced positive negative samples existing methods bot detection, reduce training abnormal model, we propose an anomaly detection framework based a combination Variational AutoEncoder algorithm. The purpose use automatically encode decode sample features. normal features more similar initial after decoding; however, there difference between decoding representation original combined, then method used detection. results show that area under curve proposed model identifying reaches 98% through experiments public datasets, which can effectively distinguish from common users further verify performance model.