A Semi-supervised Learning Method for Fake News Detection in Social Media

摘要

“Fake news” is one of the most frequent terms in news media and their spread in online social medias has been grown in recent years. Their impact affect both personal and political decisions where the latter is more important. Due to the variety of news sources and the complexity of validations, machine learning approaches are used to automatically analyze the news. The aim of this research is to detect fake news using deep learning techniques. The method is based on a semi-supervised learning framework targeting both labeled and unlabeled data using convolutional neural network. In this method, first, various features of text and image data are extracted using CNN. Then, linear discrimination analysis (LDA) is used to predict the classes of unclassified data. Also, the fitness function is modified in a way to increase the effect of estimated class in each step. Results show that the proposed method outperforms other methods in terms of recall, specificity, and sensitivity with a precision value of 95.5%.