Rumor Identification on Twitter Data for 2020 US Presidential Elections with BERT Model

Abstract

Social Media platforms provide rich resources to its users to connect, share and search the information of their interest. It is becoming part of every day’s life and politics is no different. In fact, social media platforms are becoming more significant when it comes to governmental issues and political campaigns. As information spreads within seconds, it’s extremely challenging to control and monitor the authenticity of the information. Many attempts have been made in this regard, in this paper, we briefly overview some major efforts and discuss the patterns found in the rumors and fake news that can be found by latest machine learning techniques. We extracted the tweets data specifically with hashtag_donaldtrump during the high time of 2020 US presidential election and to test their authenticity and the similar data from fact check websites Snopes.com, factcheck.org and politifact.org. We applied the already established BERT model to train on checked data and tested on the one million tweets data. In doing so, we found a reliable accuracy and proposed the fact that once all the truthful information is saved and pretrained in the model, it is able to auto identify the validation of the information shared. Also, once established such kind of models are also helpful in finding the behavior of rumors and pattern showed for American politics.