chore: move companion tables to extras and update dataset card

README.md

@@ -5,13 +5,10 @@ language:
   - en
 size_categories:
   - 10K<n<100K
-
 task_categories:
   - tabular-regression
   - tabular-classification
   - text-classification
-
-
 tags:
   - social-media
   - youtube-shorts
@@ -25,54 +22,154 @@ tags:
   - dataset
   - eda
   - music
+
+# Manual configuration so the Hub viewer/builder loads only consistent schemas
+# (and does not try to merge all CSVs with different columns).
+configs:
+  - config_name: default
+    data_files:
+      - split: train
+        path: "data/youtube_shorts_tiktok_trends_2025.csv_ML.csv"
+  - config_name: raw
+    data_files:
+      - split: train
+        path: "data/youtube_shorts_tiktok_trends_2025.csv"
 ---
 
+# 🎬 YouTube Shorts & TikTok Trends (2025)
 
-# 🎬 YouTube Shorts & TikTok Trends 2025  
-**Author:** Tarek Masryo · [Kaggle](https://www.kaggle.com/datasets/tarekmasryo/youtube-shorts-and-tiktok-trends-2025)  
-**License:** CC0 (Public Domain) — Free for open research & education  
+**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).
 
 ---
 
-## 📌 Dataset Summary  
-A **comprehensive snapshot dataset** capturing short-form video activity on **YouTube Shorts & TikTok** in **2025 (Jan–Aug)**.  
+## What’s inside
+
+This repository ships:
 
-- 🌍 Coverage: 100+ countries, 2 major platforms  
-- 📂 Package: raw video-level file, ML-ready version, monthly summaries, country rollups, top creators, top hashtags, and a data dictionary  
-- ⚡ Features: standardized schemas, deduplicated IDs, engagement metrics (views, likes, comments, shares, saves, completion rate)  
+- **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
 
-💡 For machine learning tasks, use the **ML-ready file** (`youtube_shorts_tiktok_trends_2025_ml.csv`).  
-⚠️ `trend_label` is a snapshot approximation (not full time-series). It’s a **challenging ML target** (≈25–35% baseline accuracy).  
+> Why configs? The companion tables have **different schemas**. Configs keep the Hub viewer stable and make `load_dataset()` work out of the box.
 
 ---
 
-## 🗂 Dataset Structure  
+## Dataset configs (recommended)
 
-### Main Files  
-- **`youtube_shorts_tiktok_trends_2025.csv`** → 48,079 rows × 58 columns (raw video-level data: platform, country, region, language, category, hashtags, author_handle, sound/music metadata, full engagement metrics).  
-- **`youtube_shorts_tiktok_trends_2025_ml.csv`** → 50,000 rows × 32 columns (ML-ready version: cleaned & feature-engineered for faster modeling).  
+### ✅ ML‑ready (default)
 
-### Companion Files  
-- `monthly_trends_2025.csv` → 480 rows × 8 cols (video counts, views, avg engagement, velocity)  
-- `country_platform_summary_2025.csv` → 60 rows × 14 cols (totals, medians, percentiles)  
-- `top_creators_impact_2025.csv` → 1,000 rows × 20 cols (creator stats, cumulative engagement, avg rates)  
-- `top_hashtags_2025.csv` → 82 rows × 18 cols (hashtag usage, reach, ratios, velocity)  
-- `DATA_DICTIONARY.csv` → 58 rows (column names, descriptions, types)  
+```python
+from datasets import load_dataset
 
----
+ds = load_dataset("tarekmasryo/youtube-tiktok-trends-dataset-2025")
+df = ds["train"].to_pandas()
+print(df.shape)
+print(df.head())
+```
 
-## 🚀 How to Use  
+### ✅ Raw (video-level)
 
 ```python
 from datasets import load_dataset
+
+raw = load_dataset("tarekmasryo/youtube-tiktok-trends-dataset-2025", "raw")
+raw_df = 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: platform, country, region, language, category, hashtags, creator handle, sound/music metadata, and engagement metrics.
+
+- `data/youtube_shorts_tiktok_trends_2025.csv_ML.csv`  
+  ML‑ready table: cleaned columns + feature‑engineered signals for faster modeling.
+
+> If you rename the ML file (recommended for cleanliness), also update the YAML `configs` block at the top.
+
+### 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
 
-# Load dataset from HuggingFace
-ds = load_dataset("TarekMasryo/YouTube-Shorts-TikTok-Trends-2025")
-df = ds["train"].to_pandas()
-print(df.head())
+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",
+)
 
-# Example: work with companion file
-hashtags = pd.read_csv("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.