AMAZON ML CHALLENGE · PROJECT EXPLORER

Different records. One business.

Explore how business identities connect across independent data sources.

Loading ground-truth examplesLabelled examples · no model execution

Reference businesses—

True connections—

Businesses with no matches—

Countries represented—

Explore the connections

Select a business. Inspect its matches and the records that look similar.

GROUND TRUTH EXPLORER

Select a business to begin.

Negative examples are selected for similar names. Their labels come from the provided ground truth.

BEHIND THE PROJECT

From noisy text to resolved identities.

Two retrieval channels find candidates. A matching cascade makes independent pair decisions.

Read the full architecture

THE ENGINEERING CHALLENGE

Matching beyond exact strings.

Business names, addresses, and scripts vary between sources. The project combines semantic retrieval, lexical search, and supervised matching while preserving Unicode and allowing multiple matches—or none.

24.2MSource records in the challenge dataset

1.73MSource 1 test queries in the challenge

Challenge scale, separate from the example dataset shown here.