Metric-AI/retrieval_dataset_hye

Armenian Retrieval Dataset

This dataset is part of ArmBench-TextEmbed (Armenian Embedding Model Benchmark). We manually created 185 high-quality query-document pairs covering diverse domains (finance, law, education) and use cases (customer service, sales). The annotations were reviewed by 2 independent reviewers to ensure accuracy of query-passage pairs. This dataset serves as our gold standard for Armenian retrieval evaluation.

Citation

If you use this dataset in your research, please cite our paper:

@inproceedings{navasardyan2026less,
  title={Less is More: Adapting Text Embeddings for Low-Resource Languages with Small Scale Noisy Synthetic Data},
  author={Navasardyan, Zaruhi and Bughdaryan, Spartak and Minasyan, Bagratuni and Davtyan, Hrant},
  booktitle={Proceedings of the Workshop on Low-Resource Language Models (LoResLM) at EACL 2026},
  year={2026},
  url={https://arxiv.org/abs/2603.22290}
}