Methods for Extracting and Analyzing Political Content on TikTok
Name
fadel-mdfadel-meng-bcs-2024-thesis.pdf
Description
Thesis PDF
Size
9.87 MB
Format
Adobe PDF
Checksum (MD5)
c50dc084ee29cf14b444b657f6a498ed
Author(s)
Fadel, Marie Diane
Advisor(s)
Rand, David
Date Issued
May 2024
Publisher
Massachusetts Institute of Technology
Abstract
In this thesis, I investigate the dynamics of political discourse on TikTok, with a focus on crafting a comprehensive methodology for extracting and analyzing political content related to the 2024 U.S. Presidential Election. This research utilizes a blend of advanced computational tools and crowd-sourced evaluations to delve into the mechanisms through which political influence is both exerted and perceived on the platform. For data collection, the study employed TikAPI, a tool designed for systematic scraping of TikTok videos, which targeted specific political hashtags to amass a substantial dataset. This dataset was analyzed using a variety of innovative methods, including snowball sampling to ensure a representative range of political engagement, and integration with Python to automate the data collection process. Additionally, I utilized Large Language Models (LLMs) to evaluate the relevance and persuasive impact of the content, and these machine-generated insights were then benchmarked against human judgments. Overall, the findings indicate a slight preference for Republican discourse on TikTok. Moreover, I demonstrate that OpenAI’s GPT can effectively classify videos by topic, although human input remains essential for more nuanced tasks such as stance detection and evaluation of persuasive effect. This exploration into the political landscape of TikTok represents one of the first of its kind, with the primary aim of this thesis being to develop a methodology that will support future research in this field.
MIT Department
Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
