rasyosef/amharic-passage-retrieval-dataset-with-negatives

Amharic Passage Retrieval Dataset with Negatives

This dataset is a version of amharic-news-category-classification that has been filtered, deduplicated, and formatted for passage retrieval.

This dataset can be used directly with Sentence Transformers to train Amharic Text embedding and Reranking models.

Hard Negatives:

The negative_passages column contains hard negative passages that were mined using the roberta-amharic-text-embedding-medium embedding model.

Source Datasets:

• https://huggingface.co/datasets/rasyosef/amharic-news-category-classification

Models and Code

The following Text Embedding and ColBERT late-interaction retrieval models were trained using this dataset as part of our ACL 2025 Findings paper: Optimized Text Embedding Models and Benchmarks for Amharic Passage Retrieval.

• Models Collection: https://huggingface.co/collections/rasyosef/amharic-text-embedding-models-679cb55eae1d498e3ac5bdc5
• Code: https://github.com/kidist-amde/amharic-ir-benchmarks
• Paper: https://arxiv.org/abs/2505.19356

Citation

@inproceedings{mekonnen2025amharic,
  title={Optimized Text Embedding Models and Benchmarks for Amharic Passage Retrieval},
  author={Kidist Amde Mekonnen, Yosef Worku Alemneh, Maarten de Rijke },
  booktitle={Findings of ACL},
  year={2025}
}

@misc{https://doi.org/10.48550/arxiv.2103.05639,
  doi = {10.48550/ARXIV.2103.05639},
  url = {https://arxiv.org/abs/2103.05639},
  author = {Azime, Israel Abebe and Mohammed, Nebil},
  keywords = {Computation and Language (cs.CL), Artificial Intelligence (cs.AI), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {An Amharic News Text classification Dataset},
  publisher = {arXiv},
  year = {2021},
  copyright = {arXiv.org perpetual, non-exclusive license}
}