TikTok-10M Dataset

Dataset Description

TikTok-10M is a large-scale dataset containing 10 million short-form posts from TikTok, designed for video understanding, multimodal learning, and social media content analysis. The dataset was curated to bridge the gap between academic video datasets and actual user-generated content, providing researchers with authentic patterns and characteristics of modern short-form video content that dominates social media platforms.

Request for granular highly curated data

Please fill out this form if you are training models and want additional curated datasets: Google Form

Dataset Structure

Each data instance contains:

• Post metadata (id, url, desc, challenges, create_time, etc)
• Post statistics (digg_count, comment_count, play_count, etc)
• Point of interest data (poi_name, address, poi_category, etc)
• Music data (music_name,music_album, etc)
• Video data (vq_score, duration, etc)

Usage Examples

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("The-data-company/TikTok-10M")

# Access training split
train_dataset = dataset["train"]

# Access a sample
sample = train_dataset[0]
print(f"description: {sample['desc']}")
print(f"likes: {sample['digg_count']}")
print(f"url: {sample['url']}")

Dataset Statistics

• Total Videos: 10,000,000
• Total Duration: [To be calculated]
• Average Video Length: [To be calculated]
• Category Distribution: [To be calculated]

Limitations and Biases

• Temporal Bias: Dataset reflects trending TikTok content during Spring 2025
• Geographic Bias: Dataset contains only content with US-based points of interest
• Content Bias: Dataset focuses on trending content
• Quality Variation: User-generated content varies significantly in production quality

Data Access

The dataset is available through the Hugging Face datasets library. Due to its large size (10M videos), consider using streaming or downloading specific splits based on your needs.

Ethical Considerations

• Public Data Only: Only publicly available data is included
• Privacy Protection: No personally identifiable information beyond what was publicly shared
• Content Moderation: Users should implement appropriate content filtering for their use cases
• Responsible Use: This dataset should be used in accordance with TikTok's terms of service and applicable laws

Citation

@dataset{tiktok_10m_2025,
  title={TikTok-10M: A Large-Scale Short Video Dataset for Video Understanding},
  author={The Data Company},
  year={2025},
  url={https://huggingface.co/datasets/The-data-company/TikTok-10M},
  note={A dataset of 10 million TikTok posts for multimodal learning and social media analysis}
}