README.md

---
license: cc-by-4.0
task_categories:
  - text-classification
  - video-classification
language:
  - de
tags:
  - tiktok
  - politics
  - social-media
  - germany
  - multimodal
  - content-moderation
size_categories:
  - 100K<n<1M
---

# PoliTok-DE

A Multimodal Dataset of Political TikToks and Deletions From Germany.

**Paper:** [PoliTok-DE: A Multimodal Dataset of Political TikToks and Deletions From Germany](https://arxiv.org/abs/2509.15860)

**Authors:** Tomas Ruiz, Andreas Nanz, Ursula Kristin Schmid, Carsten Schwemmer, Yannis Theocharis, Diana Rieger

## Dataset Description

This corpus contains multimodal data (video, audio, images, text) from TikTok posts related to the 2024 Saxony state election in Germany (July 1 – November 30, 2024). It includes over 195,000 posts, with more than 18,000 (17.3%) subsequently deleted from the platform.

Posts were identified through the TikTok Research API, with supplementary web scraping to obtain complete multimodal media and metadata.

## Intended Uses

- Research on intolerance and political communication patterns
- Platform deletion policies and their implications
- Multimodal analysis of political content on social media

## Dataset Access

This repository contains post IDs. Full content can be hydrated using the provided code. Access to deleted content requires approval for research purposes.

## Hydration Code

The script `hydrate.py` must be in the same directory as the `politok-de.parquet` file.
It will scrape the posts from the TikTok Website, and create for each post:

1. A video or image + audio file in the `media/` directory.
2. An entry in the `media_progress.csv` file, stating if the post was successfully scraped.
3. An entry with post metadata in the `media_metadata.csv` file, only if the post was successfully scraped.

Install dependencies:

```shell
pip install -r requirements.txt
```

Run the script:

```shell
python hydrate.py
```

Note: To interrupt the script, you might need to press `Ctrl+C` multiple times due to the concurrent nature of the scraping code.

## Citation

```bibtex
@article{ruiz2025politok,
  title={PoliTok-DE: A Multimodal Dataset of Political TikToks and Deletions From Germany},
  author={Ruiz, Tomas and Nanz, Andreas and Schmid, Ursula Kristin and Schwemmer, Carsten and Theocharis, Yannis and Rieger, Diana},
  journal={arXiv preprint arXiv:2509.15860},
  year={2025}
}
```