WELFake: Word Embedding Over Linguistic Features for Fake News Detection

Social media is a popular medium for the dissemination of real-time news all over world. Easy and quick information proliferation one reasons its popularity. An extensive number users with different age groups, gender, societal beliefs are engaged in social websites. Despite these favorable aspects, significant disadvantage comes form fake news, as people usually read share without caring about genuineness. Therefore, it imperative to research methods authentication news. To address this issue, article proposes two-phase benchmark model named WELFake based on word embedding (WE) linguistic features detection using machine learning classification. The first phase preprocesses data set validates veracity content by features. second merges feature sets WE applies voting validate approach, also carefully designs novel approximately 72 000 articles, which incorporates generate an unbiased classification output. Experimental results show that categorizes real 96.73% improves overall accuracy 1.31% compared bidirectional encoder representations from transformer (BERT) 4.25% convolutional neural network (CNN) models. Our frequency-based focused analyzing writing patterns outperforms predictive-based related works implemented Word2vec method up 1.73%.