Detecting rumors on Online Social Networks using multi-layer autoencoder

Rumors spread on Online Social Networks sometimes can lead to serious social issues. To accurately identify them from normal posts is proved to be of great value. Users' behaviors are different when they post rumors and normal posts. Since rumors only account for a small percentage of all posts, they can be regarded as anomalies. Therefore, we propose an anomaly detection method based on autoencoder to perform rumor detection. In this paper we propose several self-adapting thresholds which can facilitate rumor detection. In addition, we further discuss how the different number of hidden layers of autoencoder can influence the detection performance. The experimental results show that our model achieves a good F1 and a low false positive rate.