1.1M product reviews labeled to a 46-class defect taxonomy in seven languages.
Most review datasets stop at positive or negative. This one is labeled for what actually went wrong: a 46-class taxonomy covering sizing, battery life, shipping damage, missing parts, counterfeit suspicion, software faults and more, with a severity rating and the component at fault.
English translations sit alongside the original text in seven languages, and a double-annotated gold split is included for benchmarking. Reviews that explicitly request a return or refund are flagged, which makes the corpus useful for predicting returns as well as classifying complaints.
| Field | Type | Description |
|---|---|---|
review_id |
string | Unique identifier |
text / text_en |
string | Original review and English translation |
language |
string | ISO code, including code-mixed tags |
rating |
integer | Star rating, 1–5 |
defect_labels |
array | One or more of 46 defect classes |
severity |
string | cosmetic, functional or safety |
component |
string | Part of the product at fault |
is_return_request |
boolean | Review asks for a return or refund |
category |
string | Product category |
split |
string | train, validation or gold_test |
{
"review_id": "RV-8820113",
"text": "La batería dura poco y el cargador venía sin caja",
"text_en": "The battery does not last and the charger arrived without a box",
"language": "es",
"rating": 2,
"defect_labels": [
"battery_life_short",
"packaging_damaged"
],
"severity": "functional",
"component": "battery",
"is_return_request": false,
"category": "Audio",
"split": "train"
}
{
"review_id": "RV-8820114",
"text": "Shoe sole separated after two weeks, want a refund",
"text_en": "Shoe sole separated after two weeks, want a refund",
"language": "en",
"rating": 1,
"defect_labels": [
"material_failure",
"durability"
],
"severity": "functional",
"component": "sole",
"is_return_request": true,
"category": "Footwear",
"split": "gold_test"
}
Illustrative sample showing the structure of the data. The full dataset is delivered after purchase.
Half-hourly stock, delivery-time and price snapshots from 900 dark stores in 14 cities.
24 months of matched prices, sales rank and review velocity for 120K SKUs across six marketplaces.
Daily snapshots of 180+ hosted and open models: prices, context windows, limits and benchmark scores.
Tell us what you are trying to predict, automate or understand. We will reply within one business day with an approach, a timeline and a quote.