Someone Scraped 5.6 Billion TikTok Videos and Put the Data on Hugging Face for Free

What it claims

The dataset holds metadata for about 5.6 billion public TikTok videos from July 2014 through October 2026. It is published on Hugging Face under the datasocial account and is not gated. The page reports 1,181 downloads so far. It totals 460 GB of Parquet, split into one file per month, so you can pull specific months instead of the whole set.

Per-row fields go beyond what the summary lists:

• caption, hashtags, sound ID and on-screen text
• people mentioned in the video
• TikTok Shop product and seller IDs
• counts for views, likes, comments, shares, saves and downloads
• two of TikTok's own labels: an AI-generated flag, and whether TikTok keeps the video off the For You page

There is no creator handle column. The page says the user ID is available through datasocial.ai.

How it works

DataSocial's own write-up describes the collection method:

1. Generated device identities that pass as Android phones.
2. Reverse-engineered request signatures.
3. A spoofed TLS handshake against TikTok's private mobile API.

No login or account was used. The write-up claims the system collected 3.23 billion creator profiles, 5.94 billion videos and 2.8 billion comments in three weeks. The free release contains only the video metadata. Profiles, comments-scale data, daily updates and the scraper source ($1,699) are sold separately.

How to use it

• Licence: CC BY-NC 4.0. The dataset card says it is free for research and personal use, and sends any commercial use to datasocial.ai.
• Partitioning: monthly Parquet files suit time-sliced loading, such as recent months only.
• Sanctioned alternative: TikTok Research Tools. It requires an application and approval, and is limited to qualifying academic researchers in the US, EEA, UK, Canada and Switzerland, plus some EU not-for-profits.
• Related data: a separate 4.5-billion-video set, also described as scraped from TikTok's mobile API, is mirrored on Hugging Face.

Limits

• AI-generated flag: often empty for older videos, which come from an archive. Do not treat a missing value as "not AI-generated", especially for pre-archive-cutoff years.
• Creator analysis: without handles, you would need the paid tier for that.
• Terms of service: Section 3.4 of TikTok's US terms bars automated data extraction without written approval. The collection method described above sits squarely outside that.
• Legal risk: the page cites Reddit v. Perplexity and three scraping firms, filed October 2025. In July a federal judge largely declined to dismiss that case, and DMCA anti-circumvention claims against SerpApi survived. TikTok is not a party, and that case involves scraping Google results rather than a private API, so it is context, not precedent for this dataset.
• Other risks: training on this data, or redistributing it, carries terms-of-service exposure no matter what the CC licence says.
• Unverified figures: the scale and method claims come from the vendor's own write-up. The article does not verify them.