README.md

---
license: cc-by-sa-4.0
language:
- en
pretty_name: SemiHS-Bench
size_categories:
- 1K<n<10K
task_categories:
- text-classification
tags:
- customs-classification
- hs-code
- semiconductor
- benchmark
- dataset
---


# SemiHS-Bench


SemiHS-Bench is a semiconductor benchmark for HS6 customs classification from sparse ERP-style text.


This is the **private Hugging Face mirror** for the MIL-HAL organization. It contains the data-only release and supporting documentation. Keep this repository private until the MIL/Hitachi release decision allows public distribution.


## Contents


- 1,800 benchmark records: 900 training records and 900 evaluation records.
- 889 reference-ruling entries from CROSS, EBTI, and Japan Customs.
- HS2022 taxonomy and HS6 descriptions.
- JSON Schemas, SHA-256 manifest, statistics, datasheet, balance report, and license documentation.


The main data files are under `data/`. Evaluation, benchmark-runner, and construction code are not included.


## Canonical archival record and DOI


The canonical archival record is the Zenodo Version 2.0.0 record:


- Version DOI: https://doi.org/10.5281/zenodo.21891036
- Concept DOI for all versions: https://doi.org/10.5281/zenodo.21891035
- Zenodo record: https://zenodo.org/records/21891036


The Zenodo record is currently embargoed. This Hugging Face repository is intentionally private and should not be made public until the embargo and MIL/Hitachi release approvals are complete.


## How to cite


### Dataset citation


If you use the dataset itself, please cite the exact Zenodo release (Version 2.0.0):


Penmatsa, Tirupati Venkata Sri Rama Raju, and Swapnil Bembde. (2026). *SemiHS-Bench: A Semiconductor Benchmark for HS6 Classification from Sparse ERP-Style Text* (Version 2.0.0) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21891036


#### BibTeX


```bibtex
@misc{penmatsa2026semihsbench,
  author = {Penmatsa, Tirupati Venkata Sri Rama Raju and Bembde, Swapnil},
  title = {SemiHS-Bench: A Semiconductor Benchmark for HS6 Classification from Sparse ERP-Style Text},
  year = {2026},
  publisher = {Zenodo},
  version = {2.0.0},
  doi = {10.5281/zenodo.21891036},
  url = {https://doi.org/10.5281/zenodo.21891036},
  note = {Dataset}
}
```


### Associated publication


If you use the benchmark methodology or report experimental results, please also cite the associated CIKM 2026 paper:


Bembde, Swapnil, and Penmatsa, Sai Rama Raju. (2026). *SemiHS-Bench: A Semiconductor Benchmark for HS6 Classification from Sparse ERP-Style Text*. Paper accepted at CIKM 2026; final ACM Digital Library citation and DOI forthcoming.


Once the paper is officially published, this entry will be replaced with the final ACM citation and DOI. The Zenodo DOI above remains the correct citation for the data artifact.


## License and provenance


The dataset is released under **CC BY-SA 4.0**. See `LICENSE-DATA` for provenance and attribution requirements. The source repository is https://github.com/cti-iel/SemiHS-Bench.


The accompanying documentation notes that some source-reference and justification fields may retain supplier part numbers, names, or contact details from public rulings. Review the datasheet and `LICENSE-DATA` before any public release.