PaperlayerView exact revision on arXiv

Source-linked AI summary

Dancing to the Partisan Beat: A First Analysis of Political Communication on TikTok

Juan Carlos Medina Serrano, Orestis Papakyriakopoulos, Simon Hegelich

arXiv:2004.05478v2cs.SI

TL;DR

Political communication on TikTok had received little direct study despite the platform’s rapid growth and the rise of political videos. The paper analyzes US Republican and Democratic videos using multimodal and statistical methods to examine users, interactions, and information diffusion. It finds highly interactive communication shaped by duets, with Republicans producing more content and receiving more responses while Democrats engage more across partisan lines.

  • Problem

    The study addresses the lack of research on who uses TikTok politically and how its design shapes political information flow.

  • Method

    The authors analyze US Republican and Democratic TikTok videos using computer vision, natural language processing, statistical tools, and interaction graphs.

  • Results

    TikTok political communication is highly interactive; Republicans produced more content and received more responses, while Democrats engaged more in cross-partisan discussions.

  • Takeaways & Limitations

    TikTok’s audiovisual and duet features support political communication in which users become presenters and generate branches of responses to existing content.

  • Takeaways & Limitations

    The study’s age and gender analysis relies on image-recognition algorithms that may misclassify minorities and omit genders beyond binary male/female inferences.

Abstract

from arXiv · show

TikTok is a video-sharing social networking service, whose popularity is increasing rapidly. It was the world's second-most downloaded app in 2019. Although the platform is known for having users posting videos of themselves dancing, lip-syncing, or showcasing other talents, user-videos expressing political views have seen a recent spurt. This study aims to perform a primary evaluation of political communication on TikTok. We collect a set of US partisan Republican and Democratic videos to investigate how users communicated with each other about political issues. With the help of computer vision, natural language processing, and statistical tools, we illustrate that political communication on TikTok is much more interactive in comparison to other social media platforms, with users combining multiple information channels to spread their messages. We show that political communication takes place in the form of communication trees since users generate branches of responses to existing content. In terms of user demographics, we find that users belonging to both the US parties are young and behave similarly on the platform. However, Republican users generated more political content and their videos received more responses; on the other hand, Democratic users engaged significantly more in cross-partisan discussions.

1 INTRODUCTION

Political communication depends on platform design and users, while TikTok’s rapid growth has made it an important new setting for political discourse. The study positions TikTok against established platforms and examines how its information channels shape political communication.

  • Partisan users use social media accounts for political discourse, while much of the public obtains political news from social media.
  • Platform design configures the information channels available for political discourse, while users shape political content and a platform’s political significance.
  • TikTok became the world’s second most downloaded app in 2019 as its popularity surged.
  • Researchers have focused mainly on Facebook, Twitter, and YouTube, but users migrate between services as platform popularity changes.

Motivation

Although political behavior and platform design have been studied extensively on other social media services, TikTok had not yet received comparable analysis. This study therefore asks how partisan users, interaction structure, and information diffusion characterize political communication on TikTok.

  • Existing research has not focused on TikTok’s political users or on how its design shapes political information flow.
  • The research focuses on US politics to investigate political communication on the platform.
  • The study examines political communication on TikTok in terms of partisan users, interaction structure, and diffused content.

Original Contributions

The paper provides a first overview of partisan political communication on TikTok by analyzing multimodal content, interaction design, and user demographics. It identifies a distinctive form of political interactivity and differences in partisan activity and engagement.

  • The study provides a first overview of political communication on TikTok using videos from US Republican and Democratic partisans.
  • Computer vision, natural language processing, and statistical tools evaluate how partisans combine sound, video, and text to spread messages.
  • TikTok’s interaction design supports a novel type of political interactivity unavailable on other online social networks.
  • Republicans create more political content and receive more responses, whereas Democrats engage more with users holding opposing views.
  • The study also examines political communication alongside privacy and security issues on TikTok.

2 BACKGROUND & RELATED WORK

Political communication research has examined actors, content, news, algorithms, and platform design across major social networks, but not TikTok. TikTok’s audiovisual and recommendation features, especially duets, create distinctive ways to produce, consume, and respond to political content.

  • 2.1 Political Communication on Social Media: Prior studies examine politicians’ platform use, partisan behavior, news effects, misinformation, and algorithmic influence across social media.
  • 2.2 TikTok: TikTok had 800 million monthly active users jointly with Douyin in October 2019, and 60% of its 26.5 million US monthly users were aged 16 to 24.
  • 2.2 TikTok: TikTok videos combine text, images, video clips, sounds, voice effects, filters, and speed controls within posts up to 60 seconds long.
  • 2.2 TikTok: Background music can serve as a story builder and convey specific political messages alongside visual and textual elements.
  • 2.2 TikTok: The algorithmically generated For You page recommends videos centrally, relying less on users’ social networks than Facebook’s and Twitter’s feeds.
  • 2.2 TikTok: A duet lets users reply directly with a video displayed beside the original while preserving the original audio; reply duets can display three videos together.

3 DATA & METHODS

The study combines video collection, manual partisan coding, multimodal feature extraction, interaction-graph analysis, hashtag valence, and topic modeling to examine political communication on TikTok.

  • Data collection: 3,310 videos were collected by crawling TikTok videos containing #republican and #democrat on February 1, 2020.The authors note that hashtag search returns a limited and potentially popularity-biased set of videos.
  • Political coding: Two authors independently labeled original videos and duets as pro-Republican, pro-Democrat, or nonpartisan, with a third author resolving disagreements.Videos supporting or opposing a party or party member were classified accordingly.
  • Multimodal features: Video analysis extracted on-screen text with OCR, facial emotions, gender, and age from faces, and audio from original sounds.Videos were sampled every two seconds; processing used Tesseract, Azure Face API, and IBM text to speech.
  • Hashtag analysis: Hashtag political valence ranges from -1 for exclusively Democratic hashtags to +1 for exclusively Republican hashtags.The measure uses hashtag appearances in Democratic and Republican videos relative to each party’s total hashtags.
  • Interaction analysis: An interaction graph represented users as nodes and duet interactions as directed edges, enabling cross-ideological and intra-party comparisons.Expected interactions preserve the source user while randomly assigning the target regardless of political orientation.
  • Topic modeling: Latent Dirichlet Allocation extracted caption topics, whose Democratic and Republican video distributions were compared after selecting hyperparameters by topic coherence.Sensitivity analysis evaluated multiple topic counts and model settings.

4 RESULTS

TikTok political communication combines video-based responses, multiple audiovisual channels, and partisan content. Republican users produced more videos and attracted more interactions, while Democrats engaged more in cross-ideological exchanges.

  • Descriptive Statistics: 5,946 partisan videos comprised 2,802 original videos and 3,144 duets; Republican videos were twice as numerous and accumulated more likes, shares, and comments.Republican videos had a median of 497 likes versus 232 for Democratic videos, with significant one-sided Mann–Whitney test results for the reactions.
  • Descriptive Statistics: 70% of original videos included the user’s face, while 22% featured other video content, 7% only images, and 1% only text.For face-containing videos, the analysis extracted gender, age, and emotions from sampled frames and averaged these features per video.
  • Descriptive Statistics: Users expressed emotions similarly across parties, with happiness and surprise averaging higher than anger or sadness.The authors relate this pattern to comedy skits and sarcastic videos in which users mocked opposing political views.
  • Interaction Structure: TikTok’s duet feature adds a fourth, video-based communication level beyond viewing, liking or sharing, and commenting.Duets place the original and response videos side by side, making the interaction more similar to face-to-face communication than to a written response.
  • Interaction Structure: Duets form communication trees: original political videos generate branches of subsequent response videos.The tree begins with a political issue, followed by original content, duets to those originals, and further responses.
  • Interaction Structure: 77% of Republican duets responded to Republican users, whereas more than 80% of Democratic duets targeted Republican supporters.Observed-to-expected interaction ratios were 1.35 for Democrat–Republican exchanges and 1.28 for Republican–Republican exchanges.

5 DISCUSSION

TikTok introduces a highly interactive form of political communication in which users perform, remix, and debate political messages through audiovisual content. The study also identifies several unresolved questions about campaigning, polarization, misinformation, harmful content, privacy, and security.

  • Novel political communication: TikTok users become active presenters of political information, turning political communication into interactive audiovisual performance and politics into entertainment.Unlike platforms centered on URLs, comments, or feedback posts, TikTok users externalize political opinions through short audiovisual acts.
  • Novel political communication: Duets let users respond with reactions, fact-checking text, corrections, opposing arguments, and side-by-side comparisons of viewpoints.The authors argue that this structure most closely resembles an online public debate.
  • Open research questions: Future research should examine how TikTok profiles, political candidates, and the ban on political advertising affect campaigning and opinion formation.The study focuses on partisan content and interactions, while these additional dimensions remain to be investigated more comprehensively.
  • Open research questions: Researchers should assess whether platform design and recommendation algorithms contribute to polarization, segregation, misinformation, offensive content, or discriminatory content.The discussion also calls for research on agenda-setting effects and how controversial issues are presented.
  • Privacy and security: TikTok’s audiovisual content can expose users’ personal features and political partisanship, creating privacy and security risks for political or advertising uses.The authors note that TikTok’s viral orientation makes data readily reachable for data-mining processes.
  • Open research questions: TikTok’s open structure may support cross-partisan dialogue, but viral humor and ridicule may make political confrontation counterproductive and intensify bullying among teenagers.The authors call for psychological analyses of the platform’s influence on youth.

Ethical Concerns

The study raises ethical concerns about collecting public data involving young users and using automated systems to infer age and gender. Data deletion supported protection, but algorithmic bias and binary gender classification remain important considerations.

  • Data collection: Public TikTok data involving young users raises concerns because they may not fully understand the consequences of entering the public sphere.The researchers deleted collected materials after analysis while preserving video ids for replication.
  • Automated classification: Microsoft Azure image-recognition systems may misclassify minorities and social groups, while binary gender inference excludes other genders.The authors identify these issues as requirements for more ethical and inclusive future research.

6 CONCLUSION

This study provides a first analysis of US partisan political communication on TikTok by examining multimodal content and duet interactions. It finds distinct partisan activity and interaction patterns alongside broadly similar self-expression, while emphasizing the need for further research.

  • Study scope: The study analyzes US political videos using textual, aural, and visual information, with particular attention to TikTok’s duet feature.It compares duet interactions involving pro-Republican and pro-Democrat users.
  • Partisan differences: Republican videos were more numerous and attracted more interactions on average than Democratic videos.
  • Partisan differences: Democratic users interacted more with cross-ideological users, whereas Republican users more often generated duets from users sharing their ideology.
  • User demographics: Most users in the sample were younger than 40, and partisan users expressed themselves in similar ways despite differing political preferences.
  • Implications: Political content appears to be a relevant part of TikTok’s ecosystem, but further research is needed to understand how it is disseminated.
Loading 2004.05478v2…