Yuru Li presents project on AI parodies on TikTok at CVMR25 symposium

DCID member Yuru Li will present a project on AI-generated political parodies on TikTok together with Xin Zhou (University of Jena) at the Computational Visual Media Research (CVMR25) symposium from 5 to 7 August 2025 at the University of Zurich. 

The project uses a novel multimodal analysis method to analyse visual and narrative strategies as well as over 216,000 user comments in order to better understand the dynamics of delegitimisation. The project combines large video datasets with advanced analysis models (including ViLA, InternVL2.5) and makes a theoretical contribution to the further development of critical media analysis.

Abstract:

Al-Generated Political Parodies on TikTok: A Novel Cross-Modal Analysis of Video Content and User Comments

Xin Zhou, Yuru Li
University of Jena, University of Bremen

This project investigates how Al-generated political videos on TikTok deploy visual and narrative strategies to influence audience perception, with a particular focus on the dynamics of delegitimization. Using Ross and Rivers’ (2017) framework as an initial analytical lens, we explore four strategies-authorization, moral evaluation, rationalization, and mythopoesis-through a multimodal computational approach. Our pipeline combines large-scale video feature extraction, unsupervised clustering, semantic analysis using ViLA and InternVL2.5 models, and sentiment analysis of over 216,000 user comments. The manually curated dataset includes 1,671 GenAl 6 political videos featuring global leaders, with clustering analysis revealing seven dominant content categories such as manipulated speeches, parody role-plays, and dystopian future scenarios.
Preliminary findings show a differentiated use of delegitimization strategies across political figures, alongside distinct audience sentiment patterns. Rationalization-driven humor correlates with more positive audience reactions, while mythopoesis narratives evoke stronger negative emotions. Authorization challenges tend to polarize responses, intensifying ideological divides.
By August, I expect to present a more extensive analysis based on an expanded dataset, enabling greater generalizability across political contexts. Additionally, I anticipate offering a critical rethinking of the theoretical framework itself. Initial results suggest that while the delegitimization taxonomy provides a useful starting point, it may insufficiently capture the hybrid, affect-driven dynamics of GenAl political media. Therefore, I plan to critically evaluate and potentially replace the current framework with one better suited to computational visual media analysis-possibly integrating concepts from affect theory, visual rhetoric, or post-digital media studies.
This research thus contributes not only new empirical insights into the visual strategies of Al-mediated political communication, but also advances methodological and theoretical innovation. By bridging computational tools with critical interdisciplinary perspectives, this project aligns with CVMR’s mission to rethink how visual media can be studied at scale and critically interpreted in evolving digital ecosystems.