Geographical visualization of tweets, misinformation, and extremism during the USA 2020 presidential election using LSTM, NLP, and GIS

Abstract

Abstract Disinformation campaigns on online social networks (OSN) in recent years, have underscored democracies’ vulnerability to such operations and the importance of identifying such operations and dissecting their methods, intents, and source. With a focus on the USA 2020 presidential election, a total of 1,349,373 original Tweets have been collected by our server in real-time from the beginning of April 2020 to the end of January 2021, using four keywords: Trump, Biden, Democrats, and Republicans. In this work, deep learning, natural language processing, geographical information systems, and statistical tools are used to geographically visualize and discover if the political misinformation and extremism, political affiliation, and topics of conversations on social media are correlated with the USA 2020 presidential election results. To this end, a deep neural network is trained using 40,000 manually classified Tweets and further used to automatically classify the entire set of Tweets based on their political affiliation, topic, and whether or not they contain misinformation or extremism. It is shown that, there is a correlation between the aforementioned classes of Tweets and the election results. In other words, the political affiliation of topics and the extent of misinformation and extremism on social media are correlated with the election results to some level. The strongest correlation highlighted that the ratio of Rightist versus Leftist misinformation Tweets has a 0.67 correlation coefficient with the ratio of Trump votes versus Biden votes, across different states.

Keywords

• Misinformation
• Elections
• Social media
• Spatial visualization
• Deep learning
• Natural language processing