# Conspiracy theories and where to find them on TikTok

**Authors:** Francesco Corso, Francesco Pierri, Gianmarco De Francisci Morales  
**arXiv:** [2407.12545](https://arxiv.org/abs/2407.12545)  
**Date:** 2024-07-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2407.12545v1/

## Key points

- TikTok has gained popularity, especially among younger audiences, due to viral trends and social challenges.
- Concerns have been raised about the potential for harmful and dangerous content on the platform.
- A study using the TikTok Research API analyzed 1.5 million videos in the US over three years, finding that approximately 0.1% contained conspiracy content.
- State-of-the-art Large Language Models were used to identify harmful content with high precision but had comparable performance to traditional models like RoBERTa.
- The findings suggest the need for more effective moderation strategies and policies on social media platforms to combat malicious content.

## Summary

, , , , 

TikTok has become a popular platform, especially among younger audiences, due to its viral trends and social challenges. However, concerns have been raised about the potential for harmful and dangerous content to spread on the platform. A study utilizing the TikTok Research API collected a dataset of 1.5 million videos shared in the US over three years to analyze the presence of conspiracy theories. The study found that approximately 0.1% of all videos contained conspiracy content. The research also evaluated the impact of TikTok's Creator Program on the supply of conspiratorial content. State-of-the-art Large Language Models were used to identify harmful content in audio transcriptions with high precision, although overall performance was comparable to traditional language models like RoBERTa. The findings have significant implications for researchers and social media platforms, providing insights into developing more effective moderation strategies and policies to combat malicious content. The study also sheds light on how platform policies influence content creation trends. Limitations include potential biases in data returned by the TikTok Research API and uncertainties associated with using Large Language Models. Future research aims to address these limitations by integrating multi-modal data in classification pipelines and exploring other online media platforms. Ethical considerations were taken into account throughout the study, with measures implemented to protect user privacy and ensure data security aligns with best practices. While the dataset is not publicly available, readers can contact the authors for access to code for replication purposes. In conclusion, this study highlights the importance of understanding and mitigating harmful content on social media platforms like TikTok through advanced technologies and strategic moderation approaches.