Homegoods Taxonomy

homegoods-taxonomy is a dataset published on Hugging Face that classifies the physical goods found in an ordinary home or small business. It contains 367 classes arranged in 136 groups across 27 divisions, using a fixed-width 6-digit code scheme in which code length denotes hierarchy level. Unlike customs-oriented systems such as the Harmonised System, each class carries a physical handling attribute layer covering mass, bulk density, fragility, child-hazard status, hazmat and UN class, required handling actions, mover restrictions, and anti-affinity storage constraints, answering the question of what an object needs from its box and what it should be stored next to.

The dataset is distributed as a flat CSV file with one row per class and as a nested JSON file generated from the same YAML source. It is intended to be assigned to inventory items by a language model, so class descriptions state what falls outside each class where boundaries are unclear. Attribute values are documented as estimates accurate to roughly plus or minus 30 percent, and the README explicitly states that the dangerous-goods fields are not a shipping declaration.

The dataset is aimed at people planning household moves, those building inventory or box-level roll-ups, and developers working on storage layout or zero-shot text classification tasks. It is less suitable for retail categorisation, customs classification, insurance valuation, or any use requiring per-item accuracy.