shuttie/esci-us-small

ESCI Shopping Queries Dataset (US Locale - Small Version)

This is a curated subset of the Amazon Shopping Queries Dataset (ESCI), filtered for the US locale only and using the small version of the dataset.

Dataset Description

The Shopping Queries Dataset is a large-scale manually annotated dataset for improving product search, released by Amazon Science. It contains challenging search queries paired with products and human-labeled relevance judgments.

Original Dataset

• Total queries: ~130,000 unique queries
• Total judgments: ~2.6 million manually labeled (query, product) pairs
• Languages: English (US), Spanish (ES), Japanese (JP)

This Subset

• Locale: US (English) only
• Version: Small subset
• Format: JSONL compressed with gzip

Dataset Structure

Data Fields

• id (int): Example ID
• query (str): The search query text
• query_id (int): Unique identifier for the query
• product_id (str): Amazon product identifier (ASIN)
• product_locale (str): Product locale (always "us" in this subset)
• label (int): Relevance label (0-3, see below)
• small_version (int): Whether this example is in the small version (always 0 in this subset)
• large_version (int): Whether this example is in the large version
• split (str): Data split ("train" or "test")
• product_title (str): Product title
• product_description (str, nullable): Product description
• product_bullet_point (str, nullable): Product bullet points
• product_brand (str, nullable): Product brand name
• product_color (str, nullable): Product color
• product_locale_right (str): Product locale from join operation (always "us")

Label Mapping

The label field uses the ESCI (Exact, Substitute, Complement, Irrelevant) taxonomy:

Label	Value	Description
Exact (E)	3	The item is relevant for the query and satisfies all query specifications
Substitute (S)	2	The item is somewhat relevant but fails to fulfill some aspects of the query; can be used as a functional substitute
Complement (C)	1	The item does not fulfill the query but could be used in combination with an exact item
Irrelevant (I)	0	Products with no meaningful relation to the search query

Data Splits

• Train: Training set
• Test: Test set

Usage

Load the dataset using the HuggingFace datasets library:

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("shuttie/esci-us-small")

# Access train/test splits
train_data = dataset["train"]
test_data = dataset["test"]

# Example: Print first item
print(train_data[0])

Citation

If you use this dataset, please cite the original paper:

@article{reddy2022shopping,
  title={Shopping Queries Dataset: A Large-Scale ESCI Benchmark for Improving Product Search},
  author={Reddy, Chandan K. and others},
  journal={arXiv preprint arXiv:2206.06588},
  year={2022}
}

Dataset Creation

This subset was created by:

1. Filtering for US locale only (product_locale == "us")
2. Selecting the small version subset (small_version == 0)
3. Converting ESCI labels to integer format: E→3, S→2, C→1, I→0
4. Joining product metadata with query-product examples

Build script: build.py

License

Apache 2.0 - Same as the original ESCI dataset

Additional Information

Original Dataset Repository

amazon-science/esci-data

Related Links

• Amazon Science Dataset Page
• KDD Cup 2022 ESCI Challenge
• ArXiv Paper