The Impact of Russian Troll Tweets: Analyzing Political Motivation in Tweets from the Internet Research Agency

Abstract

This research investigates the impact of social media trolling on political events such as elections. The study utilizes the dataset of the Internet Research Agency, a Russian "troll factory" indicted by the US Justice Department in February 2018, to analyze tweets from 2012 to 2018, with the aim of creating a classification algorithm that will predict the political motivation of future tweets. The study involves running various classification algorithms, including Naive Bayes, K Nearest Neighbors, Random Forest, Histogram-based Gradient Boosting Classification Tree, and Light Gradient Boosting Machine. Data cleaning and categorization were done, assigning numeric values to each category. The resulting model scores and a breakdown of precision per category were analyzed. Word frequency analysis was also conducted to identify the most frequent words and understand the overall sentiment of the tweets and the topics targeted by the trolls. The research findings indicate that predicting the category of tweets is challenging, but the classification algorithms were able to do so with relatively good accuracy, especially for the NewsFeed and HashtagGamer categories. The study emphasizes the need for further research to improve accuracy of tweet classification algorithms and for social media platforms to continue their efforts to identify and remove troll accounts that aim to undermine political bias.