November 30, 2025
Classifying TikToks Locally: Political Content Detection with Phi-4 on Android
Can AI directly analyze political content displayed on phones without sending sensitive screen data to the cloud? This study finds that LLMs running locally can do so privately, but their current speed limitations point to hybrid methods as the most practical path for mobile media research.
Abstract
Large Language Models (LLMs) like ChatGPT-4o and Phi-4 have demonstrated great potential in identifying complex semantic content, such as political discourse, especially in cloud or desktop-based settings. However, their application on mobile devices, where privacy concerns are high, remains largely unexplored. Mobile phones have become the main way people access political discussions and news. This study investigates whether local execution of LLMs can serve as a viable, privacy-preserving way for analyzing political content seen on mobile screens. Using a Google Pixel 9, we benchmarked Phi-4 with 2,000 OCR-extracted text samples from TikTok screen recordings, comparing classification latency and feasibility. While results show that local classification is possible, latency is high, averaging over 14 seconds per sample. Although dictionary-based methods are faster, they lack the semantic flexibility of LLMs. Our findings suggest a hybrid approach and targeted frame selection strategies could enable scalable, privacy-friendly mobile media analysis in the near future.
To continue reading please visit: https://doi.org/10.1145/3771882.3773943
This open access article was published in the Proceedings of the 24th International Conference on Mobile and Ubiquitous Multimedia on 30 November 2025.
Krieter, P., de León, E., Votta, F., & Roşca, A. (2025). Classifying TikToks Locally: Political Content Detection with Phi-4 on Android. Proceedings of the 24th International Conference on Mobile and Ubiquitous Multimedia, MUM ’25, 436–438. https://doi.org/10.1145/3771882.3773943
Keywords: Political content, Human-centered computing, Ubiquitous and mobile computing systems and tools
Authors
See also
Trading nuance for scale? Platform observability and content governance under the DSA
By Charis Papaevangelou • Fabio Votta • September 22, 2025
AI transparency in political campaigns: The CampAIgn Tracker for the 2025 German Election
By Fabio Votta • February 24, 2025
What data reveals about Meta and Google’s political ad ban in the EU
By Fabio Votta • November 05, 2025
Van een verbod op TikTok naar de noodzakelijke cloudstrategie in het juridische domein
By Corien Prins • November 22, 2022
More results /
AI insiders warn it could end humanity. Does the public agree?
By Ernesto de León • Claes de Vreese • September 17, 2026
By Natali Helberger • June 16, 2026
By José van Dijck • April 30, 2026
Loss of judicial sight? The impact of assistive AI technologies on judicial perception — beyond the question of discretion
By Iris van Domselaar • Isabella Banks • September 23, 2026
By Tynke Schepers • July 09, 2026
By Corinne Cath • May 29, 2026
By Martijn Logtenberg • November 20, 2025
By Jacqueline Kernahan • Richard Bartels • Mark de Reuver • Daniel Oberski • Roel Dobbe • October 20, 2025
The Label Paradox: Can AI transparency create more distrust?
By Natali Helberger • September 21, 2026
By Jin Wan • Theo Araujo • Natali Helberger • Claes de Vreese • September 17, 2026
By João Pedro Quintais • September 03, 2026
