lxe/receipt-household-lines-odbl

Receipt Household & Grocery Lines (ODbL)

Synthetic US receipt line items built from real products in the Open Food Facts family of databases, each with a POS-style printed label, a readable product name and one of 39 spending categories (taxonomy v2, taxonomy.json). This is the share-alike part of the Receipt Kitten item-namer training data. The permissive part (real receipts, USDA-based synthetic lines, teacher-generated household inventory) is lxe/receipt-line-items-taxonomy (MIT).

config	rows	upstream
household_catalog	9,641 (Open Beauty Facts 5,623, Open Products Facts 2,755, Open Pet Food Facts 1,263)	6,023 US/Canada products exported 2026-09-28
synthetic_off_lines	7,470	Open Food Facts US products (export used to build a product lookup DB, 2026-09-27)

Fields

store, store_type, label (printed), name, category (taxonomy-v2 leaf), top_level, confidence. household_catalog adds generator, upstream_database, barcode (the product's code in that database), chain, item_id, variant (two printed labels per product), generation_confidence. synthetic_off_lines adds generator, upstream, name_v1/category_v1 (draft before relabelling), relabelled, split.

How it was built

• household_catalog: for each product with an English name sold in the US or Canada, DeepSeek V4.1 Flash (deepseek-flash) decided whether it is a real retail item, wrote a clean name and category, and invented two printed labels in the style of specific chains. A second, blind teacher pass re-read each label with only the store as context; rows were kept only when it agreed on the category, was not low-confidence and did not rename the product.
• synthetic_off_lines: product names from Open Food Facts were abbreviated by a rule-based POS-label generator (store-brand codes, vowel dropping, sizes, printer-width truncation), then re-labelled by the same teacher.
• Rows whose printed label equals one in a private evaluation set were removed.

Limitations

Labels are model-generated, and printed labels are synthetic (plausible, not observed). Open Food Facts product data is crowd-sourced and can be incomplete or wrong. Categories on boundary items (candles, light bulbs, contact-lens solution) are the least consistent.

License and attribution

This dataset is a Derivative Database of Open Food Facts, Open Products Facts, Open Beauty Facts and Open Pet Food Facts and is made available under the Open Database License (ODbL) v1.0: https://opendatacommons.org/licenses/odbl/1-0/. Individual contents are under the Database Contents License (DbCL) v1.0: https://opendatacommons.org/licenses/dbcl/1-0/. If you publicly use or redistribute it, or a database derived from it, keep this notice and share the derived database under ODbL. A model trained on it is a Produced Work: ship an attribution notice with it.

Contains information from Open Food Facts (https://world.openfoodfacts.org), Open Products Facts (https://world.openproductsfacts.org), Open Beauty Facts (https://world.openbeautyfacts.org) and Open Pet Food Facts (https://world.openpetfoodfacts.org), which is made available under the Open Database License (ODbL) v1.0. Credit: Open Food Facts contributors, Open Products Facts contributors, Open Beauty Facts contributors, Open Pet Food Facts contributors.

Names, categories and printed labels were generated with the DeepSeek Open Platform API; DeepSeek's terms assign any rights in outputs to the user.