SciMDIX Dataset

Dataset Description

SciMDIX is a bilingual dataset containing scientific abstracts in Russian and Kazakh across four domains: IT, Linguistics, Medicine, and Psychology. It is designed for advanced Information Extraction tasks and is divided into two main configurations:

1. ner_re: Contains annotations for Named Entity Recognition (NER) and Relation Extraction (RE).
2. aspects: Contains aspect-level markup (AIM, MATERIAL, METHOD, RESULT, TASK, TOOL, USAGE) for the same texts.

Data Structure

Each configuration contains four splits: train_ru, train_kz, test_ru, and test_kz. Note: You can identify the domain of a specific text by looking at the prefix in the filename column (it-, ling-, med-, psy-).

Configuration: ner_re

• filename: Original text file name (includes domain prefix).
• abstract: The raw text of the scientific abstract.
• annotation: Text with inline BRAT-style markup [Entity|ID|TYPE].
• entities: Extracted entities in BRAT format.
• relations: Extracted relations between entities.

Configuration: aspects

• filename: Original text file name (includes domain prefix).
• abstract: The raw text of the scientific abstract.
• aspect_annotation: Text with inline aspect markup [Span|ID|TYPE].
• aspects: Extracted aspects in BRAT format.

How to use

You can load the dataset using the datasets library. Specify the configuration (ner_re or aspects) and the split you want to use:

from datasets import load_dataset

# Load the Russian Train split for NER and Relation Extraction
ds_ner_ru_train = load_dataset("tvbat/SciMDIX", "ner_re", split="train_ru")
print(ds_ner_ru_train[0])

# Load the Kazakh Test split for Aspects
ds_asp_kz_test = load_dataset("tvbat/SciMDIX", "aspects", split="test_kz")

Repository

The code, models, and additional resources related to this dataset can be found in our GitHub repository

Citation

If you use the SciMDIX dataset in your research, please cite our papers:

1. Batura T., Yerimbetova A., Mukazhanov N., Shvarts N., Sakenov B., Turdalyuly M. Information Extraction from Multi-Domain Scientific Documents: Methods and Insights. Applied Sciences. MDPI. 2025. V.15, 9086.

@article{scimdix2025,
  author = {Batura, Tatiana and Yerimbetova, Aigerim and Mukazhanov, Nurzhan and Shvarts, Nikita and Sakenov, Bakzhan and Turdalyuly, Mussa},
  title = {Information Extraction from Multi-Domain Scientific Documents: Methods and Insights},
  journal = {Applied Sciences},
  volume = {15},
  year = {2025},
  number = {16},
  article-number = {9086},
  publisher = {MDPI},
  doi = {https://doi.org/10.3390/app15169086}
}

2. Shvarts N., Batura T., Mukazhanov N., Yerimbetova A., Turdalyuly M., Sakenov B. SciMDIX: A dataset for aspect extraction from multi-domain scientific documents in Kazakh and Russian. Procedia Computer Science. 2026. V. 275, pp.474-483.

@article{scimdix2026,
  title={SciMDIX: A dataset for aspect extraction from multi-domain scientific documents in Kazakh and Russian},
  author={Shvarts, Nikita and Batura, Tatiana and Mukazhanov, Nurzhan and Yerimbetova, Aigerim and Turdalyuly, Mussa and Sakenov, Bakzhan},
  journal={Procedia Computer Science},
  volume={275},
  pages={474--483},
  year={2026},
  publisher={Elsevier},
  doi = {https://doi.org/10.1016/j.procs.2026.01.056}
}