GravelAI/beauty-sample-sunscreens-uk

Gravel AI — Beauty Product Sample Dataset (Sunscreens × UK)

A free evaluation slice of the Gravel AI Raw Data Service: every sunscreen product tracked on UK retail shelves, entity-resolved and INCI-parsed, with canonical trend tags.

Snapshot: September 2026 · Licence: free for evaluation & research with attribution (see LICENSE.txt)

Dataset structure

Three tables joined on product_id:

Config	Rows	Description
products	2,617	One row per entity-resolved sunscreen on UK shelves — brand, name, first-seen month, rating, review count, description
ingredients	91,096	Parsed INCI lists — one row per (product, label position, ingredient)
trend_tags	83,122	Canonical trend tags per product — trend name, angle (ingredient / format / claims), launch month

471 brands represented.

products.csv

Column	Type	Description
product_id	string	Stable cross-retailer product ID (Gravel entity resolution). Join key.
brand	string	Resolved brand name, harmonised across retailers and spellings
product_name	string	Product display name (English)
category	string	Gravel category slug — sunscreens in this sample
country	string	Market — UK in this sample
first_seen_month	string YYYY-MM	First month observed on shelf
star_rating	float	Average consumer rating (0–5) where exposed by the source
num_reviews	int	Review count where exposed by the source
description	string	Product marketing description (English)

ingredients.csv

Column	Type	Description
product_id	string	Joins to products
inci_position	int	1-based position on the INCI label. Position correlates with concentration — EU rules order ingredients ≥1% by descending concentration.
ingredient	string	Ingredient name as parsed from the INCI list

trend_tags.csv

Column	Type	Description
product_id	string	Joins to products
canonical_trend	string	Gravel canonical trend name (e.g. "Long-Duration Hydration")
angle	string	Trend lens: ingredient, formulation (format), claims, and category-specific angles
launch_month	string YYYY-MM	Month the observation is attributed to

Why this dataset exists

It demonstrates the three layers that separate structured beauty data from scraped listings:

1. Entity resolution — one record per real product, deduplicated across retailers. product_id is stable across monthly refreshes.
2. INCI parsing — ingredient lists as ordered, structured rows rather than raw label strings.
3. Trend tagging — every product mapped to a canonical trend taxonomy.

Example usage

from datasets import load_dataset
import pandas as pd

products = load_dataset("GravelAI/beauty-sample-sunscreens-uk", "products")["train"].to_pandas()
inci     = load_dataset("GravelAI/beauty-sample-sunscreens-uk", "ingredients")["train"].to_pandas()
tags     = load_dataset("GravelAI/beauty-sample-sunscreens-uk", "trend_tags")["train"].to_pandas()

# Most-used ingredients in UK sunscreens
print(inci["ingredient"].str.lower().value_counts().head(20))

# Hero ingredients only (top 5 INCI positions)
heroes = inci[inci.inci_position <= 5]
print(heroes["ingredient"].str.lower().value_counts().head(20))

# Launch cadence by month
print(products.groupby("first_seen_month").size())

What the full service adds

This is 1 category × 1 country. The full Raw Data Service covers 6 million products from 45,000 brands across 40 categories and 200+ retailers in 30+ countries, refreshed monthly, adding pricing and availability history, trade-name / supplier attribution, monthly deltas keyed on stable IDs, and delivery via CSV/Parquet, Snowflake, BigQuery, S3 or REST API — with commercial AI-training licences.

→ gravelai.com/raw-data · Free sample page · Book a demo

Excluded from the sample

Retailer identities and URLs, pricing history, trade-name/supplier attribution, trend evidence text, other categories and countries, and monthly deltas — all reserved for the commercial tiers.

Citation

@misc{gravelai2026beautysample,
  title  = {Gravel AI Beauty Product Sample Dataset (Sunscreens × UK)},
  author = {Gravel AI},
  year   = {2026},
  url    = {https://gravelai.com/raw-data/sample}
}

Questions / full-dataset access: hello@gravelai.com