Datasets:

jmhb
/

BioASQ

@@ -46,7 +46,7 @@ This dataset is derived from the **BioASQ Challenge** data. BioASQ is a series o
 ### Citations
-If you use this dataset, please cite the following papers:
 **Primary BioASQ Paper**:
 ```bibtex
@@ -62,25 +62,16 @@ If you use this dataset, please cite the following papers:
 }
 ```
-**If using this organized version**:
-```bibtex
-@misc{bioasq_all_types,
-  title={BioASQ All Question Types Dataset},
-  author={[Your Name]},
-  year={2026},
-  note={Reorganized version of BioASQ challenge data with question types as splits},
-  url={https://huggingface.co/datasets/jmhb/BioASQ}
-}
-```
 **PaperSearchQA** (if used in context of this project):
 ```bibtex
-@inproceedings{papersearchqa2026,
-  title={PaperSearchQA: A Large-Scale Dataset for Question Answering over Biomedical Literature},
-  author={[Authors to be added]},
-  booktitle={Proceedings of the 2026 Conference of the European Chapter of the Association for Computational Linguistics (EACL)},
-  year={2026},
-  url={https://jmhb0.github.io/PaperSearchQA}
 }
 ```
@@ -99,196 +90,55 @@ For full license terms, see: https://bioasq.org/participate
 ### This Dataset
-This reorganized version is provided **AS-IS** for research purposes and is subject to the original BioASQ terms. The reorganization (splitting by question type) is released under **CC BY 4.0**, but the underlying data remains under BioASQ terms.
 ## Dataset Structure
-### Splits
-The dataset has **four splits**, one for each question type:
-| Split | Samples | Description |
-|-------|---------|-------------|
-| `factoid` | 1,609 | Questions with short factual answers (e.g., "What is the function of protein X?") |
-| `yesno` | 1,464 | Yes/No questions (e.g., "Is aspirin effective for headaches?") |
-| `summary` | 1,283 | Questions requiring summarization (e.g., "Describe the role of...") |
-| `list` | 1,048 | Questions with list-based answers (e.g., "List all known variants of...") |
-### Fields
-Each sample contains the following fields:
-```python
-{
-    'id': str,                    # Unique question ID
-    'type': str,                  # Question type ('factoid', 'yesno', 'summary', or 'list')
-    'question': str,              # The question text
-    'answer': list or str,        # Answer(s) - format varies by type
-    'ideal_answer': str or list,  # Detailed/ideal answer
-    'documents': list,            # Relevant PubMed document IDs (URLs)
-    'snippets': list,             # Relevant text snippets from documents
-    'concepts': list,             # Medical concepts mentioned
-    'triples': list,              # Knowledge graph triples (if available)
-    'asq_challenge': str,         # BioASQ challenge identifier
-    'folder_name': str,           # Source folder
-}
-```
-### Answer Formats by Type
-- **factoid**: `answer` is a list of short text answers
-- **yesno**: `answer` is "yes" or "no"
-- **summary**: `answer` is typically empty; use `ideal_answer` for summary text
-- **list**: `answer` is a list of items
 ## Usage
-### Load All Splits
 ```python
 from datasets import load_dataset
-# Load entire dataset
-dataset = load_dataset("jmhb/BioASQ_all_types")
-print(dataset.keys())
-# dict_keys(['factoid', 'yesno', 'summary', 'list'])
-# Access specific split
-factoid_questions = dataset['factoid']
-print(f"Factoid questions: {len(factoid_questions)}")
-```
-### Load Specific Split
-```python
 # Load only factoid questions
-factoid_data = load_dataset("jmhb/BioASQ_all_types", split="factoid")
-# Example question
-sample = factoid_data[0]
-print(f"Question: {sample['question']}")
-print(f"Answer: {sample['answer']}")
-print(f"Documents: {sample['documents'][:2]}...")
-```
-### Filter by Type
-```python
-# Get all yes/no questions
-yesno_data = load_dataset("jmhb/BioASQ_all_types", split="yesno")
-# Count yes vs no answers
-yes_count = sum(1 for ex in yesno_data if ex['answer'] == 'yes')
-no_count = sum(1 for ex in yesno_data if ex['answer'] == 'no')
-print(f"Yes: {yes_count}, No: {no_count}")
-```
-### Iterate Over All Types
-```python
-dataset = load_dataset("jmhb/BioASQ_all_types")
-for question_type, split_data in dataset.items():
-    print(f"\n{question_type.upper()} Questions:")
-    print(f"  Total: {len(split_data)}")
-    # Show first example
-    example = split_data[0]
-    print(f"  Sample: {example['question'][:100]}...")
-```
-## Data Statistics
-### Overall Statistics
-- **Total Questions**: 5,404
-- **Question Types**: 4 (factoid, yesno, summary, list)
-- **Source**: BioASQ challenges 1-9
-- **Domain**: Biomedical and life sciences
-- **Language**: English
-### Question Type Distribution
 ```
-factoid:  29.8% (1,609 questions)
-yesno:    27.1% (1,464 questions)
-summary:  23.7% (1,283 questions)
-list:     19.4% (1,048 questions)
-```
-### Sample Questions by Type
-**Factoid**:
-- "What is the genetic basis of Huntington's disease?"
-- "Which protein is encoded by the BRCA1 gene?"
-**Yes/No**:
-- "Is the protein Papilin secreted?"
-- "Does metformin interfere with vitamin B12 absorption?"
-**Summary**:
-- "Describe the role of the immune system in cancer development."
-- "What is known about the association between coffee consumption and health?"
-**List**:
-- "List symptoms of Alzheimer's disease."
-- "Which genes are associated with autosomal dominant Alzheimer's disease?"
-## Limitations and Considerations
-1. **Question Distribution**: Not uniformly distributed across types
-2. **Answer Variability**: Answer formats vary significantly by type
-3. **Domain Specificity**: Highly specialized biomedical knowledge required
-4. **Evaluation Complexity**: Different metrics needed for different question types
-5. **Document Access**: Referenced PubMed documents may require separate retrieval
-6. **Knowledge Cutoff**: Questions are based on medical knowledge available up to the challenge date
-## Evaluation
-Different question types require different evaluation metrics:
-- **Factoid**: Exact Match (EM), F1 score
-- **Yes/No**: Accuracy, F1 score
-- **Summary**: ROUGE, BERTScore, human evaluation
-- **List**: F1 score, Partial Match
-See the [BioASQ evaluation tools](http://participants-area.bioasq.org/general_information/Task6b/) for official evaluation scripts.
-## Related Resources
-### BioASQ
-- **Website**: https://bioasq.org/
-- **Participate**: https://bioasq.org/participate
-- **Papers**: https://bioasq.org/publications
-### PubMed Corpus
-- **Full PubMed Corpus**: https://huggingface.co/datasets/jmhb/pubmed_bioasq_2022
-- **NLM PubMed**: https://pubmed.ncbi.nlm.nih.gov/
-### PaperSearchQA Project
-- **Website**: https://jmhb0.github.io/PaperSearchQA
-- **GitHub**: https://github.com/jmhb0/PaperSearchQA
-- **Main Dataset**: https://huggingface.co/datasets/jmhb/PaperSearchQA
-- **Collection**: https://huggingface.co/collections/jmhb/papersearchqa
-## Contact and Support
-For questions about:
-- **Original BioASQ data**: Contact BioASQ organizers at https://bioasq.org/contact
-- **This dataset organization**: Open an issue on the PaperSearchQA GitHub
-- **PaperSearchQA project**: Visit https://jmhb0.github.io/PaperSearchQA
 ## Acknowledgments
-- **BioASQ Organizers**: For creating and maintaining this valuable resource
-- **PubMed/NCBI**: For the underlying biomedical literature
-- **Challenge Participants**: For advancing biomedical QA research
-- **Funding Agencies**: Supporting the BioASQ challenges
-## Version History
-- **v1.0** (2026-01): Initial release with all question types as separate splits
-  - 5,404 questions across 4 types
-  - Organized from BioASQ challenges 1-9
-  - Split by question type for convenient access

 ### Citations
+If you use this dataset, please cite:
 **Primary BioASQ Paper**:
 ```bibtex
 }
 ```
 **PaperSearchQA** (if used in context of this project):
 ```bibtex
+@misc{burgess2026papersearchqalearningsearchreason,
+      title={PaperSearchQA: Learning to Search and Reason over Scientific Papers with RLVR},
+      author={James Burgess and Jan N. Hansen and Duo Peng and Yuhui Zhang and Alejandro Lozano and Min Woo Sun and Emma Lundberg and Serena Yeung-Levy},
+      year={2026},
+      eprint={2601.18207},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2601.18207},
 }
 ```
 ### This Dataset
+This reorganized version (with question types as splits) is provided for research convenience. All terms and conditions of the original BioASQ license apply.
 ## Dataset Structure
+### Data Fields
+Each sample contains:
+- `id`: Unique question identifier
+- `body`: The question text
+- `type`: Question type (factoid, yesno, summary, or list)
+- `ideal_answer`: Reference answer(s)
+- `exact_answer`: Structured answer (for factoid/list questions)
+- `documents`: URLs of relevant PubMed documents
+- `snippets`: Relevant text snippets from the documents
+### Data Splits
+The dataset is organized into 4 splits by question type:
+| Split | Examples |
+|-------|----------|
+| factoid | 1,609 |
+| yesno | 1,464 |
+| summary | 1,283 |
+| list | 1,048 |
+| **Total** | **5,404** |
 ## Usage
 ```python
 from datasets import load_dataset
+# Load all question types
+dataset = load_dataset("jmhb/BioASQ")
 # Load only factoid questions
+factoid = load_dataset("jmhb/BioASQ", split="factoid")
+# Load specific question types
+dataset = load_dataset("jmhb/BioASQ", split=["factoid", "yesno"])
 ```
+## Additional Resources
+- **BioASQ Official Website**: https://bioasq.org/
+- **BioASQ Participants Area**: https://participants-area.bioasq.org/
+- **PaperSearchQA Project**: https://jmhb0.github.io/PaperSearchQA/
 ## Acknowledgments
+This dataset is made available through the BioASQ Challenge organizers. We thank them for creating and maintaining this valuable resource for the biomedical NLP community.