Review of Feature Extraction Techniques for Fake News Detection

Abstract

The widespread use of social media channel, as well as the development of these networks, has offered a platform for fake news to propagate quickly among people. For society, social media may be a two-edged weapon, serving as either a simple avenue for exchanging ideas or an unforeseen conduit for disseminating fake news to a broad audience. With the rise of social media, individuals may now distribute information for free, with hardly any inquiry and fewer restrictions than earlier. Detecting fake news has become a major concern. Document classification can be aided by extracting features. The technique of extracting a set of significant features is known as feature extraction from supervised and unsupervised data to help with categorization. It is critical to correctly identify the text's relevant features. Many artificial intelligent models have been used to achieve cutting-edge outcomes in Natural Language Processing applications. Using online available datasets, the performance of various feature extraction algorithms is compared. Mainly the review is focused on NLP in which text preprocessing includes tokenization, stopwords removal, and lemmatization. Different techniques for extracting features like Term Frequency- Inverse Document Frequency (TFIDF) Vectorizer with Ngram analysis vectorization, Term Frequency- Inverse Document Frequency (TFIDF) Vectorizer with Ngram analysis vectorization, Bag of Words, and approaches for Word Embeddings like Word2vec, BERT, FastText, and GloVe are compared.