from datasets import load_dataset

comments = load_dataset("tiagozip/tiktok", "comments", split="train", streaming=True)
videos = load_dataset("tiagozip/tiktok", "videos", split="train", streaming=True)

format

config	rows	notes
comments	3.2B+	one file per month, `data/comments/YYYY-MM.parquet`
videos	23M+	one file per month the video was posted
repost_edges	54M+	who reposted which video, by month first seen
users	300k+	users whose reposts were crawled

comments

column	type	notes
`cid`	uint64	comment id
`aweme_id`	uint64	video id, joins to `videos.id`
`reply_id`	uint64	0 when not a reply
`reply_to_reply_id`	uint64	0 when not a reply to a reply
`text`	string
`create_time`	datetime	UTC
`digg_count`	int64	likes, counter at fetch time
`reply_comment_total`	int64	same
`comment_language`	string	TikTok's guess
`author_pin`	bool	pinned by the video author
`is_author_digged`	bool	liked by the video author
`user_hash`	uint64	keyed hash of the commenter, not their id
`mentions`	list[string]
`hashtags`	list[string]
`fetched_at`	datetime

videos

id, author_sec_uid, author_unique_id, author_uid, description, create_time, duration, digg_count, comment_count, play_count, share_count, collect_count, music_id, music_title, challenges, text_language, is_ad, fetched_at.

repost_edges and users

reposter_hash and sec_hash / uid_hash use the same keyed hash as comments.user_hash, so they join to each other but not back to TikTok accounts.

what this is not

not complete. comments were crawled per video, so this covers the videos that were crawled, not all of TikTok.
engagement counts are snapshots. every counter is whatever it read at fetched_at.

license

by downloading the data, you agree to the license terms. tl;dr:

no republishing, sharing, or selling the data (linking here is fine)
no using it, or publishing anything made with it, without clear and obvious credit
no machine learning, AI training, or publishing embeddings and similar representations
research and publication are allowed, with credit and approval
you must honour opt-outs and deletions

disclaimers

access can be revoked at any time and all data is provided as-is. this dataset contains personal information about real people, a lot of whom are not public figures. this dataset is not a reflection of the views of the author or the institutions they are affiliated with. please be respectful. every comment in this dataset was publicly readable at the time it was fetched.

TikTok and associated marks are trademarks of their respective owners. this dataset is independent and unaffiliated.

opting out

email legal@tiago.zip with your username and your data is removed from the next update.

citation

@misc{tiktok,
  title  = {tiktok: a corpus of public TikTok comments},
  author = {tiagozip},
  year   = {2026},
  url    = {https://huggingface.co/datasets/tiagozip/tiktok}
}

Downloads last month: 46

Total file size:

208 GB