DETECTING EXTREMISM IN TWEETS USING LEXICON-BASED TECHNIQUE

ABSTRACT

Social media enhances communication across geopolitical boundaries. While it serves as a tool for healthy interactions, it has also been used to propagate disinformation aimed at causing fear, disaffection, and other acts that could lead to physical attacks. This use of computers and the internet to promote such extremist views is considered cyberterrorism. This study proposes an unsupervised learning technique to detect extremism in tweets and identify those who make such posts. A lexicon-based technique was used to compute sentiment scores of tweets, and a rule-based method was applied to categorise them. Experiments were carried out with tweets posted before and after the Kuje prison break of July 2022 in Abuja, Nigeria. Correlation analysis shows that tweet engagements have strong relationships. Also, it was noted that negative tweets tend to spread faster, with extremely negative tweets getting the most engagement, such as likes and retweets. Results from the experiment showed that the framework produced 0.90 accuracy and recall, 0.96 precision, and the F1 score was 0.93. Future studies may consider other machine learning tools and improve accuracy in automatically detecting extremism on Tweets.

Keywords: Twitter, Extremism, Cyberterrorism, Sentiment Analysis, Kuje Prison Break