A Corpus-Based Sentiment Analysis of AI-Generated Deepfakes in Political Discourse

Abstract

This study examines how Pakistani social media users perceive AI-generated deepfakes in political discourse, using a corpus-based approach to analyze responses to synthetic videos of leaders like Nawaz Sharif and Imran Khan. Based on Framing Theory (Goffman, 1974; Entman, 1993), the research draws from 20,000–30,000 Instagram comments across 10 posts (January–June 2025). Sentiment analysis, employing LIWC and Apilayer tools, identifies a spectrum of emotional responses, including joy, anger, sarcastic joy, amused disbelief, disdain, mockery, scornful amusement, confusion, bewilderment, excitement, thrill, and amused anticipation, revealing the nuanced emotional landscape of user reactions. Combined with qualitative framing analysis, the study finds that deepfakes heighten polarization and distrust in Pakistan's divided political context. The research advances methods for multilingual digital analysis and extends framing theory to a political setting. It suggests practical steps, such as media literacy programs and platform rules, to reduce misinformation risks. Limitations include a focus on urban users, with future work recommended for broader, longitudinal studies.