README
TikTok Videos: 5.6 billion
5,597,462,038 public TikTok videos (2014 → Oct 2026): captions, hashtags, sounds, TikTok Shop links and
engagement stats. One Parquet file per month in data/. Snapshot of 2026-10-03.
Need more? → datasocial.ai
Creator profiles, sounds, daily history, a live database updated every day, and the scraper code.
Query it
DuckDB (reads only the columns it needs):
SELECT hashtag, count(*) AS videos
FROM (SELECT unnest(hashtags) AS hashtag FROM 'hf://datasets/datasocial/tiktok-5.6B-videos/data/2026-09.parquet')
GROUP BY hashtag ORDER BY videos DESC LIMIT 20;
Python:
import pandas as pd
df = pd.read_parquet("hf://datasets/datasocial/tiktok-5.6B-videos/data/2026-09.parquet",
columns=["video_id", "posted_at", "views", "caption"])
Columns
| Column | Type | Notes |
|---|---|---|
video_id | UInt64 | tiktok.com/@_/video/{video_id} |
posted_at | DateTime | UTC |
country | String | region TikTok gives the video |
language | String | un = undetermined |
type | String | video or photo |
duration_ms | UInt32 | 0 on photos |
image_count | UInt8 | photo posts |
width, height | UInt16 | pixels |
caption | String | |
hashtags, hashtag_ids | Array | lower-case, parallel arrays |
mentions | Array(UInt64) | creator_ids |
on_screen_text | Array(String) | text stickers |
sound_id | UInt64 | 0 = none |
is_ad | UInt8 | Spark Ads |
branded_content | String | paid_partnership, shop_affiliate or empty |
product_id, seller_id | UInt64 | TikTok Shop; 0 = none |
is_shop_video | UInt8 | |
is_ai_generated | UInt8, nullable | null = TikTok did not say |
not_recommended_for_fyp | UInt8, nullable | TikTok keeps it off the For You page |
sound_muted | UInt8 | audio removed for copyright |
views, likes, comments, shares, saves, downloads | UInt64 | as of stats_updated_at |
engagement_rate | Float32, nullable | (likes+comments+shares+saves) / views |
stats_updated_at | DateTime | UTC |
Older videos come from an archive: some fields are 0 or null there (size, downloads, AI flag), and duration is rounded to the second.
License
CC BY-NC 4.0: free for research and personal use. Commercial use, creator profiles and daily updates: datasocial.ai.
Similar repos
README
TikTok Videos: 5.6 billion
5,597,462,038 public TikTok videos (2014 → Oct 2026): captions, hashtags, sounds, TikTok Shop links and
engagement stats. One Parquet file per month in data/. Snapshot of 2026-10-03.
Need more? → datasocial.ai
Creator profiles, sounds, daily history, a live database updated every day, and the scraper code.
Query it
DuckDB (reads only the columns it needs):
SELECT hashtag, count(*) AS videos
FROM (SELECT unnest(hashtags) AS hashtag FROM 'hf://datasets/datasocial/tiktok-5.6B-videos/data/2026-09.parquet')
GROUP BY hashtag ORDER BY videos DESC LIMIT 20;
Python:
import pandas as pd
df = pd.read_parquet("hf://datasets/datasocial/tiktok-5.6B-videos/data/2026-09.parquet",
columns=["video_id", "posted_at", "views", "caption"])
Columns
| Column | Type | Notes |
|---|---|---|
video_id | UInt64 | tiktok.com/@_/video/{video_id} |
posted_at | DateTime | UTC |
country | String | region TikTok gives the video |
language | String | un = undetermined |
type | String | video or photo |
duration_ms | UInt32 | 0 on photos |
image_count | UInt8 | photo posts |
width, height | UInt16 | pixels |
caption | String | |
hashtags, hashtag_ids | Array | lower-case, parallel arrays |
mentions | Array(UInt64) | creator_ids |
on_screen_text | Array(String) | text stickers |
sound_id | UInt64 | 0 = none |
is_ad | UInt8 | Spark Ads |
branded_content | String | paid_partnership, shop_affiliate or empty |
product_id, seller_id | UInt64 | TikTok Shop; 0 = none |
is_shop_video | UInt8 | |
is_ai_generated | UInt8, nullable | null = TikTok did not say |
not_recommended_for_fyp | UInt8, nullable | TikTok keeps it off the For You page |
sound_muted | UInt8 | audio removed for copyright |
views, likes, comments, shares, saves, downloads | UInt64 | as of stats_updated_at |
engagement_rate | Float32, nullable | (likes+comments+shares+saves) / views |
stats_updated_at | DateTime | UTC |
Older videos come from an archive: some fields are 0 or null there (size, downloads, AI flag), and duration is rounded to the second.
License
CC BY-NC 4.0: free for research and personal use. Commercial use, creator profiles and daily updates: datasocial.ai.