Abstract

The automatic detection of rumors in social networks is an important problem that would allow counteracting the effects that the propagation of false information produces. We study the performance of deep learning architectures in this problem, analyzing ten different machines on word2vec and BERT. Our results show that some architectures are more suitable for some particular classes, suggesting that the use of committee machines would offer advantages in this task.