An Empirical Analysis of Fake Tweet Detection Using Statistical and Deep Learning Approaches

Abstract

In today’s era, fake news has become a rising threat. Fake news could be defined as news which involves deliberate lies or flaws, spread via traditional news media. The spread and propagation of such fake news poses significant threats to the individual as well as the national security. Moreover, the authenticity of social media article is never been verified frequently. A variety of approaches have been proposed so far with respect to fake news recognition, to avoid the victims of deceptive social media. This paper presented a data-driven perspective using natural language processing method, machine learning and deep learning approaches to detect fake tweets. We have analyzed the performance of various statistical classifier algorithms like logistics regression, random forest and Naïve Bayes with different performance features like count vectorizer and tf-idf. Also, we analyzed the performance of long short-term memory (LSTM) model, convolutional neural network and hybrid approach to improve the overall accuracy.