Skip to content

Dataset · tasksource

Product search query relevance benchmark

tasksource/esci

A large dataset from Amazon with query-product pairs and relevance judgments in three languages to improve search.

What's inside

It contains text and tabular data for evaluating product search relevance: combining product descriptions and relevance labels (ESCI). The database contains from 1 to 10 million records in English, Japanese, and Spanish. The source data is taken from the Amazon Science research benchmark and pre-cleaned.

What you can do with it

  • Training and evaluating text-retrieval ranking models for e-commerce search
  • Fine-tuning query and product card relevance classifiers
  • Evaluating the quality of multilingual embeddings in search tasks
Tasks
Text classification, Text retrieval
Languages
English, Japanese, Spanish
Size
1M–10M rows
Downloads this month
83.1K
Likes
10

How to load it

from datasets import load_dataset

ds = load_dataset("tasksource/esci")