Machine Learning-Based Analysis of Anti-U.S. Perception on Social Media

The goal of this project is to investigate the mass media and social media effect in creating anti-U.S. perceptions through misinformation and disinformation efforts within the U.S. allies. The target country for the case study is Turkey, and the actors are Russia and USA. The social media we are analyzing is Twitter. We captured Twitter data using Twitter’s Filtered stream API by providing specific Turkish keywords related to the U.S., and we automatically translated collected tweets into English and applied sentiment and emotion analysis techniques on social media data to identify the stance for or against the U.S. We visualized a time series of the number of Turkish tweets related to the USA from January 2021 to September 2021. We also visualized time series of the sentiments and emotions of the tweets. We saw from peaks in our time series that there are important political events associated with them. Currently, we are working on classifying users into different groups such as bots, public users, government officials and media agencies to gain more insight into misinformation and disinformation. This will be done by identifying patterns in tweeting, the volume of posts from one user, the time difference between consecutive posts of one user, and so on. Our future steps are data collection from 2015 and onward and performing sentiment and emotion analysis of all tweets. The long-term goal is to become a research hub for compiling data about anti-American views regarding disinformation and misinformation in Turkey and possibly other countries.

Hello. My name is Gowri Prathap, and I am a senior majoring in Computational and Data Sciences. I work with Dr. Hamdi Kavak as part of this research project funded by the Commonwealth Cyber Initiative. Other partner institutions in this project are Old Dominion University and Tidewater Community College. The goal of this project is to investigate the mass media and social media effect in creating anti-U.S. perceptions through misinformation and disinformation efforts within the U.S. allies. The target country for the case study is Turkey, and the actors are Russia and USA. The social media we are analyzing is Twitter, and the mass media we are analyzing are Turkey’s local and international media outlets publishing in Turkish. This presentation summarizes our efforts in the last three months.

I am part of the team as an Undergraduate Research Assistant, and I am working on the social media aspect of the project, which is Twitter. We captured Twitter data using Twitter’s Filtered stream API by providing specific Turkish keywords related to the U.S., and we automatically translated collected tweets into English and applied sentiment and emotion analysis techniques on social media data to identify the stance for or against the U.S. Sentiment analysis is the process of identifying whether the sentiment of a tweet is positive, negative, or neutral. Each tweet is assigned a score depicting its sentiment score. We used various Python packages to conduct sentiment analysis, such as Afinn, Textblob and Vader, which give the sentiment score of a tweet.

Emotion analysis identifies emotions in a tweet such as happiness, anger, fear, disgust, and so on. We performed this using the Pysentimiento and Text2Emotion packages in Python.

Here we visualize a time series of the number of Turkish tweets related to the USA from January 2021 to September 2021. There is a total of 3526900 tweets from January to September. On an average day, the number of tweets per day is about 10000. However, we can see peaks in our data. The peak we see on January 20th was because the inauguration of Joe Biden took place on that day. The peak we see on April 24 is because Biden recognized the Armenian Genocide on that day. On June 14th there was a meeting between Turkey president Erdogan and Biden, and on August 16 Taliban took over Afghanistan. So, we can see that with the peaks, there are important political events associated. The Old Dominion team is working towards gathering manually collected events between U.S. and Turkey to qualitatively analyze sentiment changes.

This is the sentiment score time series, and similarly, we can see the drop in sentiment when Biden recognized the Armenian Genocide and the peak in sentiment when Erdogan and Biden met. Here, we have a time series of the emotions in the tweets. We can see that disgust is the most prevalent in Turkish tweets about the USA, followed by Anger and Fear. When Biden recognized the Armenian Genocide, there was a surge in the fear and disgust emotions.

Currently, we are working on classifying users into different groups and gaining more insight into misinformation and disinformation. I am aiming to identify and classify users as bots, public users, government officials, media agencies, etc. This will be done by identifying patterns in tweeting, the volume of posts from one user, the time difference between consecutive posts of one user, etc. Here, we can see a plot of users who have posted less than 50 posts from Mid 2020 to Mid 2021. This is the plot for the number of users who have posted more than 50 posts. User classification is in progress, and we are currently working on it. Our future steps are data collection from 2015 and onward and performing sentiment and emotion analysis of all tweets. We are working on Twitter user categorization and comparing different Twitter user categories. Our long-term goal is to become a research hub for compiling data about anti-American views regarding disinformation and misinformation in Turkey and possibly other countries. Thank you for listening, and please drop your questions below.