bemir / README.md
s-emanuilov's picture
Update README.md
1675956 verified
metadata
configs:
  - config_name: corpus
    data_files:
      - split: bg
        path: corpus_bg.jsonl
      - split: en
        path: corpus_en.jsonl
  - config_name: queries
    data_files:
      - split: bg
        path: queries_bg.jsonl
      - split: en
        path: queries_en.jsonl
  - config_name: qrels
    data_files:
      - split: train
        path: qrels/train.jsonl
      - split: dev
        path: qrels/dev.jsonl
      - split: test
        path: qrels/test.jsonl
task_categories:
  - text-retrieval
language:
  - bg
  - en
pretty_name: bemir
license: cc-by-4.0
tags:
  - medical
size_categories:
  - 1K<n<10K

BEMIR Dataset

Bulgarian-English Medical Information Retrieval

📊 Dataset Overview

Metric Count Details
Documents 7,266 3,633 BG + 3,633 EN
Queries 6,474 3,237 BG + 3,237 EN
Relevance Pairs 134,294 Train: 110,575, Dev: 11,385, Test: 12,334

📁 Files & Sizes

File Size Records Description
corpus_bg.jsonl 11 MB 3,633 Bulgarian documents
corpus_en.jsonl 5.7 MB 3,633 English documents
queries_bg.jsonl 315 KB 3,237 Bulgarian queries
queries_en.jsonl 218 KB 3,237 English queries
qrels/train.jsonl 7.2 MB 110,575 Training relevance
qrels/dev.jsonl 756 KB 11,385 Development relevance
qrels/test.jsonl 819 KB 12,334 Test relevance

📝 Content Statistics

Documents

Language Count Avg Length Sample
Bulgarian 3,633 1,594 chars (289 words) Скорошни проучвания предполагат, че статини, утвърдена група лекарства за превенция на сърдечно-съдо...
English 3,633 1,496 chars (272 words) Recent studies have suggested that statins, an established drug group in the prevention of cardiovas...

Queries

Language Count Avg Length Examples
Bulgarian 3,237 28 chars (5 words) • Клетките на рака на гърдата се хранят с холестерол
• Диета за лечение на астма и екзема
English 3,237 21 chars (3 words) • Breast Cancer Cells Feed on Cholesterol
• Using Diet to Treat Asthma and Eczema

🎯 Relevance Judgments

Split Query-Doc Pairs Unique Queries Unique Documents Avg Pairs/Query
Train 110,575 2,590 3,588 42.7
Dev 11,385 324 3,119 35.1
Test 12,334 323 3,128 38.2

🔧 Data Format

// Corpus
{"id": "0406436662", "title": "Document title...", "text": "Document content...", "lang": "en"}

// Queries  
{"id": "df3aceec36", "text": "Query text...", "lang": "en"}

// QRels
{"query-id": "df3aceec36", "corpus-id": "0406436662", "score": "1"}

📚 Citation

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

@dataset{emanuilov2025bemir,
  title={BEMIR: Bulgarian-English Medical Information Retrieval Dataset},
  author={Simeon Emanuilov},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/llm-bg/bemir}
}