Firmansyah-Ibrahim/indo-bloom-raw-bse

📚 Indo-Bloom BSE RAW Corpus

<div style="background-color: #e6f7ff; padding: 15px; border-left: 5px solid #1890ff; border-radius: 5px; margin-bottom: 20px;"> <strong>⚠️ RESEARCH ARTIFACT STATUS: RAW CORPUS (Stage 0)</strong><br><br> This dataset serves as the <strong>raw material corpus</strong> for the Indo-Bloom research project at <strong>Universitas Negeri Malang (UM)</strong>.<br><br> <strong>Current State:</strong> <code>Extracted & Cleaned Context from BSE Textbooks</code><br> <strong>Next Stage:</strong> <code>QA Pair Generation (Stage 1) → Silver Corpus</code> </div>

<div style="background-color: #fff3cd; padding: 12px; border-left: 5px solid #ffc107; border-radius: 5px; margin-bottom: 20px;"> <strong>🔒 FROZEN — Raw v1.0</strong><br> This version is <strong>permanently frozen</strong> to ensure reproducibility. This corpus will be used as input for QA generation pipeline. </div>

📄 Associated Publication

Preprint Paper: Ibrahim, F., Prasetya, D. D., & Widiyaningtyas, T. (2026). Towards Indo-Bloom: A Preliminary Study on Controllable Bloom's Taxonomy-aligned Question Generation in Indonesian. TechRxiv.

DOI: `10.36227/techrxiv.177220024.48567886/v1` Publication Date: February 27, 2026 Status: 📝 Preprint (Not Peer-Reviewed)

Read the Full Paper: TechRxiv Preprint

🎓 Dissertation Details

• —Project Title: A Unified Framework for Controllable Indonesian Automatic Question Generation Aligned with Bloom's Taxonomy
• —Principal Investigator: Firmansyah Ibrahim (Ph.D. Candidate)
• —Student ID (NIM): 250534903885
• —Institution: Doctoral Program in Electrical and Informatics Engineering, Universitas Negeri Malang (UM)
• —Supervision Team:
• —Promotor: Dr. Eng. Didik Dwi Prasetya, S.T., M.T.
• —Co-Promotor: Dr. Ir. Triyanna Widiyaningtyas, M.T.

📝 Dataset Description

Indo-Bloom BSE RAW Corpus is a curated collection of educational text chunks extracted from Indonesian high school textbooks (Buku Sekolah Elektronik - BSE) published by Kemdikbudristek. This corpus serves as the foundational raw material for generating Bloom's Taxonomy-aligned question-answer pairs in the Indo-Bloom research project.

Purpose in Research Pipeline

This dataset represents Stage 0 in the Indo-Bloom methodology:

Stage 0 (This Dataset) → Stage 1 (QA Generation) → Stage 2 (Expert Annotation) → Stage 3 (Gold Standard)

Role:

• —Provides clean, domain-specific context passages for QA generation
• —Ensures educational content quality and curriculum alignment
• —Enables controlled generation of pedagogically valid questions

📊 Dataset Specifications

1. Corpus Statistics

2. Subject Distribution

3. Grade Level Distribution

📂 Data Structure

Each row in the dataset contains the following fields:

💻 How to Use

Loading the Dataset (Python)

from datasets import load_dataset

# Load the BSE RAW corpus
dataset = load_dataset("Firmansyah-Ibrahim/indo-bloom-bse-raw")

# Print first example
print(dataset['train'][0])

# Output structure:
# {
#   'chunk_id': 'chunk_0001',
#   'mata_pelajaran': 'Sejarah',
#   'jenjang': 'SMA/MA',
#   'kelas': 'X',
#   'context': '...',
#   'word_count': 160,
#   'noise_score': 0,
#   'source_file': 'Sejarah_X_2024.pdf',
#   'page_range': '15-20'
# }

Filtering by Subject

import pandas as pd

# Load as pandas DataFrame
df = pd.DataFrame(dataset['train'])

# Filter by subject
sejarah_chunks = df[df['mata_pelajaran'] == 'Sejarah']
geografi_chunks = df[df['mata_pelajaran'] == 'Geografi']
biologi_chunks = df[df['mata_pelajaran'] == 'Biologi']
sosiologi_chunks = df[df['mata_pelajaran'] == 'Sosiologi']

print(f"Sejarah: {len(sejarah_chunks)} chunks")
print(f"Geografi: {len(geografi_chunks)} chunks")

Filtering by Grade Level

# Filter by grade
kelas_10 = df[df['kelas'] == 'X']
kelas_11 = df[df['kelas'] == 'XI']
kelas_12 = df[df['kelas'] == 'XII']

print(f"Kelas X: {len(kelas_10)} chunks")
print(f"Kelas XI: {len(kelas_11)} chunks")
print(f"Kelas XII: {len(kelas_12)} chunks")

Quality Filtering

# Get only high-quality chunks (noise_score = 0)
clean_chunks = df[df['noise_score'] == 0]

# Filter by word count range
medium_chunks = df[(df['word_count'] >= 100) & (df['word_count'] <= 200)]

print(f"Clean chunks: {len(clean_chunks)}")
print(f"Medium-length chunks: {len(medium_chunks)}")

🔧 Data Processing Pipeline

Extraction Process (IBEX v3.0)

graph TD
    A[BSE PDF Files] -->|IBEX Extractor| B[Raw Text Extraction]
    B -->|Noise Filter L1| C[Remove Instructional Content]
    C -->|Noise Filter L2| D[Remove Exercise/Quiz Patterns]
    D -->|Sentence Cleaning| E[Remove Pilgan Patterns]
    E -->|Chunking Algorithm| F[Fixed-size Chunks 150±50 words]
    F -->|Quality Check| G[Noise Score = 0]
    G -->|Metadata Tagging| H[Final BSE RAW Corpus]

Technical Specifications

Extraction Tool

• —Tool: IBEX v3.0 (Indo-Bloom Context Extractor)
• —PDF Library: PyMuPDF (fitz)
• —Processing: Python 3.8+

Cleaning Filters

Level 1 (Noise Patterns):

• —Instructional phrases: "Simak gambar", "Jawablah pertanyaan"
• —Learning objectives: "Tujuan pembelajaran", "Kata kunci"
• —Metadata: ISBN, author names, page numbers

Level 2 (Sentence Cleaning):

• —Multiple-choice patterns: a. ... b. ... c. ...
• —Question prompts: "Diskusikan dengan teman"
• —Activity instructions: "Buatlah laporan"

Chunking Algorithm

# Chunking strategy
CHUNK_SIZE = 150  # words
OVERLAP = 37      # 25% overlap
MIN_LENGTH = 50   # minimum viable chunk
MAX_LENGTH = 450  # maximum to prevent context overflow

# Quality threshold
NOISE_THRESHOLD = 0  # Only chunks with score 0 included

🔄 Dataset Relationship in Research Pipeline

Stage 0: BSE RAW Corpus (This Dataset)

Input: PDF textbooks from Kemdikbudristek Process: IBEX v3.0 extraction & cleaning Output: 1,825 clean educational text chunks Status: ✅ Completed & Frozen

Stage 1: QA Generation

Input: BSE RAW Corpus (this dataset) Process: Qwen2.5-3B-Instruct with Kemdikbud prompts Output: Indo-Bloom Silver Corpus (2,768 QA pairs) Status: ✅ Completed

Stage 2: Expert Annotation

Input: Indo-Bloom Silver Corpus Process: Human expert validation (3 annotators) Output: Indo-Bloom Annotated Corpus Status: 🔵 In Progress

Stage 3: Gold Standard

Input: Indo-Bloom Annotated Corpus Process: IAA validation (Kappa ≥ 0.70) Output: Indo-Bloom Gold Corpus v1.0 Status: 🟡 Planned

📋 Extraction Scripts Documentation

Script: IBEX v3.0 (Indo-Bloom Context Extractor)

Input: BSE PDF files Output: CSV with cleaned text chunks

Key Features:

• —Margin removal (8% top/bottom)
• —Hyphenation correction
• —Noise pattern detection (30+ patterns)
• —Chunk size balancing with overlap
• —Domain metadata preservation

Extraction Statistics:

🚀 Research Roadmap

• —[x] Stage 0 (Completed): BSE text extraction & cleaning (This Dataset)
• —[x] Stage 1 (Completed): QA generation with LLM (Silver Corpus)
• —[ ] Stage 2 (In Progress): Expert annotation
• —[ ] Stage 3 (Planned): IAA validation & Gold Standard release
• —[ ] Stage 4 (Future): Model training & evaluation

⚖️ License & Attribution

Dataset License

License: CC BY 4.0

You are free to:

• —✅ Share — copy and redistribute the material
• —✅ Adapt — remix, transform, and build upon the material

Under the following terms:

• —Attribution — You must give appropriate credit

Source Material

Original Content: Buku Sekolah Elektronik (BSE) Publisher: Kementerian Pendidikan, Kebudayaan, Riset, dan Teknologi (Kemdikbudristek) Indonesia Original License: CC BY 4.0 (as per BSE repository)

Books Included:

• —Sejarah Indonesia (Kelas X, XI, XII)
• —Geografi Indonesia (Kelas X, XI, XII)
• —Biologi (Kelas X, XI, XII)
• —Sosiologi (Kelas X, XI, XII)

📜 Citation

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

Primary Citation (Dataset Repository)

@misc{ibrahim2026bseraw,
  author       = {Ibrahim, Firmansyah},
  title        = {Indo-Bloom BSE RAW Corpus: Educational Context Extraction from Indonesian Textbooks},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/Firmansyah-Ibrahim/indo-bloom-bse-raw}},
  note         = {Raw corpus for Indo-Bloom AQG research. Extracted from BSE Kemdikbudristek.}
}

Secondary Citation (Associated Research Paper)

@article{ibrahim2026indobloom_paper,
  author    = {Ibrahim, Firmansyah and Prasetya, Didik Dwi and Widiyaningtyas, Triyanna},
  title     = {Towards Indo-Bloom: A Preliminary Study on Controllable Bloom's 
               Taxonomy-aligned Question Generation in Indonesian},
  journal   = {TechRxiv},
  year      = {2026},
  month     = {February},
  day       = {27},
  doi       = {10.36227/techrxiv.177220024.48567886/v1},
  url       = {https://doi.org/10.36227/techrxiv.177220024.48567886/v1},
  note      = {Preprint — not peer-reviewed}
}

Combined Citation (For Academic Papers)

Text Format:

Ibrahim, F. (2026). Indo-Bloom BSE RAW Corpus: Educational Context Extraction from Indonesian Textbooks. Hugging Face. https://huggingface.co/datasets/Firmansyah-Ibrahim/indo-bloom-bse-raw Associated paper: Ibrahim, F., Prasetya, D. D., & Widiyaningtyas, T. (2026). Towards Indo-Bloom. TechRxiv. https://doi.org/10.36227/techrxiv.177220024.48567886/v1

📞 Contact & Support

Principal Investigator: Firmansyah Ibrahim (Ph.D. Candidate) 📧 Email: firmansyah.ibrahim.2505349@students.um.ac.id 🏛️ Institution: Universitas Negeri Malang 🔗 Dataset: indo-bloom-bse-raw 📄 Paper: TechRxiv Preprint

Supervision Team:

• —Dr. Eng. Didik Dwi Prasetya, S.T., M.T. — Promotor
• —Dr. Ir. Triyanna Widiyaningtyas, M.T. — Co-Promotor

🙏 Acknowledgments

We gratefully acknowledge:

• —Kemdikbudristek Indonesia for publishing BSE textbooks under CC BY 4.0
• —Universitas Negeri Malang for institutional support
• —Hugging Face for open-source hosting infrastructure
• —Indonesian NLP Community for domain expertise
• —TechRxiv (IEEE) for preprint hosting

⚠️ Disclaimer

This dataset is extracted from educational materials published by Kemdikbudristek Indonesia. While we have performed careful cleaning and quality control, users should:

• —✅ Verify extracted content against original sources for critical applications
• —✅ Respect the original CC BY 4.0 license terms
• —✅ Acknowledge both this corpus and the original BSE sources

Educational Use: This corpus is intended for research and educational purposes, particularly for developing AI-assisted educational tools aligned with the Indonesian national curriculum.

🔗 Related Datasets

Last Updated: February 28, 2026 Version: Raw v1.0 — 🔒 FROZEN Status: ✅ Production Ready — Stage 0 Complete Paper DOI: `10.36227/techrxiv.177220024.48567886/v1`

© 2026 Firmansyah Ibrahim | Universitas Negeri Malang