---
pretty_name: "YouTube Shorts & TikTok Trends (2025)"
license: cc-by-4.0
language:
  - en
size_categories:
  - 10K<n<100K
task_categories:
  - tabular-regression
  - tabular-classification
  - text-classification
tags:
  - social-media
  - youtube-shorts
  - tiktok
  - short-form-video
  - trends
  - engagement
  - content-intelligence
  - time-series
  - tabular
  - dataset
  - eda
  - music

# Manual configuration so the Hub viewer/builder loads only consistent schemas
# (and does not try to merge multiple CSVs with different columns).
configs:
  - config_name: default
    data_files:
      - split: train
        path: "data/youtube_shorts_tiktok_trends_2025_ml.csv"
  - config_name: raw
    data_files:
      - split: train
        path: "data/youtube_shorts_tiktok_trends_2025.csv"
---

# 🎬 YouTube Shorts & TikTok Trends (2025)

**Author:** Tarek Masryo  
**License:** CC BY 4.0

A structured snapshot of short-form video activity across **YouTube Shorts** and **TikTok** during **2025 (Jan–Aug)**.  
Built for **content intelligence**, **analytics dashboards**, and **ML baselines** (classification/regression).

---

## What’s inside

This repository ships:

- **Two loadable dataset configs** (via `datasets.load_dataset`):
  - `default` → **ML-ready** table (cleaned + modeling-friendly)
  - `raw` → **raw video-level** table (wider schema + richer metadata)
- **Companion tables** under `extras/` (aggregations + dictionary):
  - monthly rollups, country/platform summaries, creator and hashtag leaderboards, and a data dictionary

`load_dataset(repo)` loads the **default** config automatically.

> Why configs? The companion tables have **different schemas**. Configs keep the Hub viewer stable and make `load_dataset()` work out of the box.

---

## Quickstart (recommended)

### ✅ ML-ready (default)

```python
from datasets import load_dataset

repo = "tarekmasryo/youtube-tiktok-trends-dataset-2025"
df = load_dataset(repo)["train"].to_pandas()
print(df.shape)
print(df.head())
```

### ✅ Raw (video-level)

```python
from datasets import load_dataset

repo = "tarekmasryo/youtube-tiktok-trends-dataset-2025"
raw_df = load_dataset(repo, "raw")["train"].to_pandas()
print(raw_df.shape)
print(raw_df.head())
```

---

## Repository structure

### Main tables (Hub loadable)

- `data/youtube_shorts_tiktok_trends_2025.csv`  
  Raw video-level data (wider schema + richer metadata): platform, country, region, language, category, hashtags, creator handle, sound/music metadata, and engagement metrics.

- `data/youtube_shorts_tiktok_trends_2025_ml.csv`  
  ML-ready table: cleaned columns + feature-engineered signals for faster modeling.

### Companion tables (download as files)

These are not part of the dataset builder; they are supporting assets:

- `extras/monthly_trends_2025.csv` → monthly aggregates (counts, views, engagement/velocity proxies)
- `extras/country_platform_summary_2025.csv` → country/platform rollups (totals, medians, percentiles)
- `extras/top_creators_impact_2025.csv` → creator-level stats (reach, cumulative engagement, rates)
- `extras/top_hashtags_2025.csv` → hashtag usage, reach, ratios, velocity signals
- `extras/DATA_DICTIONARY.csv` → column names, descriptions, and types

---

## Download companion tables (works anywhere)

```python
import pandas as pd
from huggingface_hub import hf_hub_download

repo_id = "tarekmasryo/youtube-tiktok-trends-dataset-2025"
path = hf_hub_download(
    repo_id=repo_id,
    repo_type="dataset",
    filename="extras/top_hashtags_2025.csv",
)

hashtags = pd.read_csv(path)
print(hashtags.sort_values("views", ascending=False).head(10))
```

---

## Typical use cases

### Content intelligence & dashboards
- Platform comparison (Shorts vs TikTok)
- Country / region distribution and concentration
- Category performance and seasonality (Jan–Aug)
- “High-opportunity” segments (coverage vs velocity)

### ML baselines
- **Classification:** trend labels / buckets (if present)
- **Regression:** engagement proxies (views, likes, completion rate, etc.)
- **Text tasks:** titles/hashtags/sounds (if present) for weak supervision or clustering

---

## Notes on labels & evaluation

- If `trend_label` exists, treat it as a **snapshot approximation**, not ground-truth time-series labeling.
- Report your:
  - split strategy (time-aware vs random)
  - leakage checks (creator leakage, sound leakage, near-duplicate videos)
  - baseline metrics (not just best score)

---

## Limitations & responsible use

- This dataset covers **Jan–Aug 2025** and should not be treated as full historical coverage.
- Engagement signals are **platform-dependent** and affected by sampling, availability, and regional conventions.
- If any identifiers exist (creator handles, video IDs), avoid doxxing or targeting individuals; use aggregated insights where possible.

---

## Citation

If you use this dataset in research or public work, cite it as:

**Tarek Masryo. (2025). _YouTube Shorts & TikTok Trends (2025)_. Hugging Face Datasets.**

---

## Changelog

See `CHANGELOG.md` for version notes.