As we gradually open up access to the BONSAI database, it has become clear that one of the main challenges is how to link external data to BONSAI. Each user may rely on their own taxonomy of products or items, and mapping these to align with BONSAI can be time-consuming and complex. This process is often a major bottleneck for interoperability.
To help address this, we are making part of that work openly available. We have published the BONSAI Product Classification Mappings on Zenodo, a curated collection of more than 10,000 correspondence records linking products from a wide range of external classifications to the BONSAI product classification.
The dataset has been developed as part of BONSAI (https://bonsai.uno), a not-for-profit initiative to build an open and transparent sustainability database. These mappings are used internally to integrate heterogeneous data sources into BONSAI, but we believe they can also benefit the wider community.
By publishing these mappings, we aim to:
The current release contains mappings covering classifications such as:
Each correspondence links a source product code or label to one or more BONSAI products, together with available provenance information.
Because different classification systems describe products at different levels of detail, the mappings are not always one-to-one. A single source product may correspond to multiple BONSAI products, and multiple source products may map to the same BONSAI product.
We also see this dataset as a useful benchmark for AI training.
Large language models and retrieval systems are becoming increasingly capable of understanding product descriptions, but they still require high-quality labelled examples. The published mappings provide real-world training and evaluation data for developing automated product retrieval systems that can effectively interpret BONSAI’s classification.
This first release is just the beginning. If you discover errors, have additional correspondence tables, or would like to contribute mappings from other classifications, we would be very interested in collaborating.
Our long-term vision is to build an open, community-maintained ecosystem of product mappings that lowers the barrier to integrating sustainability data from diverse sources.
Zenodo dataset: https://doi.org/10.5281/zenodo.21297037
