Word embedding and classification methods and their effects on fake news detection

Abstract

Natural language processing contains multiple methods of translating written text or spoken words into numerical information called word embeddings. Some of these embedding methods, such as Bag of Words, assume words are independent of one another. Other embedding methods, such as Bidirectional Encoder Representations from Transformers and Word2Vec, capture the relationship between words in various ways. In this paper, we are interested in comparing methods treating words as independent and methods capturing the relationship between words by looking at the effect these methods have on the classification of fake news. Using various classification methods, we compare the word embedding processes based on their effects on accuracy, precision, sensitivity, and specificity.