mfaruqi/indonesian-medical-ner

Indonesian Medical Complaint NER Dataset

Overview

The Indonesian Medical Complaint NER Dataset is a manually curated Named Entity Recognition (NER) dataset consisting of Indonesian patient complaints collected from publicly available health discussion forums. The dataset is designed to support the development of Indonesian clinical NLP applications, particularly for extracting structured medical information from unstructured patient complaints.

The dataset contains annotations for five medical entity types:

• GEJALA (Symptoms)
• PENYAKIT (Diseases or Medical Conditions)
• OBAT (Medications or Supplements)
• TINDAKAN (Medical Procedures or Treatments)
• LOKASI (Affected Body Parts or Anatomical Locations)

---

Dataset Statistics

Split	Number of Samples
Train	1,418
Validation	304
Test	305
Total	2,027

Entity Distribution (BIO Labels)

Label	Count
B-GEJALA	1,583
I-GEJALA	707
B-PENYAKIT	767
I-PENYAKIT	418
B-OBAT	224
I-OBAT	152
B-LOKASI	334
I-LOKASI	92
B-TINDAKAN	297
I-TINDAKAN	209
O	53,005

---

Annotation Schema

GEJALA (Symptoms)

Patient-reported symptoms or complaints.

Examples:

• nyeri
• mual
• demam
• telinga berdengung
• vagina kering

---

PENYAKIT (Diseases)

Diagnosed diseases or medical conditions.

Examples:

• diabetes
• hipertensi
• endometriosis
• PCOS

---

OBAT (Medications)

Medications, supplements, or therapeutic products.

Examples:

• paracetamol
• ibuprofen
• omega-3
• antibiotik

---

TINDAKAN (Medical Procedures)

Medical procedures, examinations, or interventions.

Examples:

• operasi
• biopsi
• USG
• kemoterapi

---

LOKASI (Body Locations)

Body parts or anatomical locations affected by symptoms or diseases.

Examples:

• perut
• perut bagian bawah
• kulit kepala
• tulang rusuk
• dagu

---

Data Format

Each sample is stored in JSON format:

{
  "text": "Dok, saya sering merasa nyeri di perut bagian bawah.",
  "annotations": [
    {
      "token": "nyeri",
      "label": "GEJALA"
    },
    {
      "token": "perut bagian bawah",
      "label": "LOKASI"
    }
  ],
  "url": "https://example.com"
}

---

Example

Input

Dok, saya mengalami nyeri di perut bagian bawah dan mual sejak kemarin.

Output

{
  "GEJALA": [
    "nyeri",
    "mual"
  ],
  "LOKASI": [
    "perut bagian bawah"
  ]
}

---

Intended Use

This dataset can be used for:

• Named Entity Recognition (NER)
• Medical Complaint Structuring
• Information Extraction from Patient Complaints
• Clinical NLP Research for Indonesian Language
• Medical Chatbots
• Retrieval-Augmented Generation (RAG)
• Symptom Mining and Healthcare Analytics

---

Annotation Process

The dataset was created through the following steps:

1. Collection of publicly available Indonesian medical complaint texts.
2. Initial annotation using Large Language Models (LLMs).
3. Manual verification and correction of all annotations.
4. Conversion into BIO tagging format for NER training.

---

Baseline Model

A baseline NER model was trained using:

• Model: indobenchmark/indobert-base-p1
• Task: Token Classification
• Labels: BIO scheme with 5 entity types
• Evaluation Metric: Entity-level F1 Score (SeqEval)

Best model performance:

Metric	Score
Precision	0.6278
Recall	0.7240
F1 Score	0.6725

---

Limitations

• The dataset is relatively small compared to large-scale biomedical corpora.
• The data mainly consists of informal patient complaints in Indonesian.
• The dataset is not intended for medical diagnosis or clinical decision-making.

---

Citation

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

@dataset{faruqi2026indonesianmedicalcomplaintner,
  title={Indonesian Medical Complaint NER Dataset},
  author={Muhammad Faruqi},
  year={2026},
  publisher={Hugging Face}
}

---

License

Please ensure compliance with the terms of use of the original data sources. This dataset is intended solely for research and educational purposes.

---

Contact

Muhammad Faruqi

GitHub: https://github.com/mfaruqi35

Hugging Face: https://huggingface.co/mfaruqi