An Unsupervised Bayesian Neural Network for Truth Discovery in Social Networks

The problem of estimating event truths from conflicting agent opinions in a social network is investigated. An autoencoder learns the complex relationships between truths, reliabilities and observations. A Bayesian model proposed to guide learning process by modeling relationship autoencoder's outputs with different variables. At same time, it also models agents network. approach unsupervised applicable when ground truth labels events are unavailable. variational inference method used jointly estimate hidden variables parameters autoencoder. Experiments on three real datasets demonstrate that our competitive with, most cases better than, several state-of-the-art benchmark methods.