tekwhisperer/Cannabis_Science_Data

---
license: apache-2.0
task_categories:
- question-answering
- text-generation
language:
- en
tags:
- chemistry
- biology
- medical
- cannabis
- research
- cannabisextraction
- plant
- extraction
- chemicalengineering
- synthetic-data
- scientific-qa
pretty_name: Cannabis-Science-Literature
size_categories:
- 100K<n<1M
---

# Cannabis Science Literature QA Dataset

This dataset contains **161,170 high-quality question-answer pairs** derived from over 400 peer-reviewed cannabis science research papers and textbooks. Created to advance AI research in cannabis science and medical applications, it provides a comprehensive resource for training language models on cannabis-related scientific knowledge.

## Dataset Details

### Dataset Description

This dataset was systematically generated from a curated collection of cannabis science literature using advanced NLP processing techniques. The source materials include peer-reviewed research papers, academic journals, and college-level textbooks covering cannabis chemistry, biology, pharmacology, extraction methods, and medical applications.

- **Curated by:** Kellan Finney
- **Funded by:** Eighth Revolution
- **Language(s) (NLP):** English
- **Total Q&A Pairs:** 161,170
- **Source Documents:** 400+ research papers and textbooks
- **License:** Apache 2.0

### Dataset Sources

- **Repository:** [https://github.com/KellanFinney/Canna_LoRA](https://github.com/KellanFinney/Canna_LoRA)
- **Source Papers:** [Cannabis Research Literature Collection](https://drive.google.com/drive/folders/1zOrIlrChpPteq7cmeNluCBA6tquwIvj9?usp=drive_link)

## Uses

### Direct Use

- **Training scientific Q&A models** for cannabis domain expertise
- **Fine-tuning language models** for cannabis and botanical applications
- **Research applications** in computational biology and chemistry
- **Educational chatbots** for cannabis science learning
- **Literature analysis** and knowledge synthesis tools

### Out-of-Scope Use

- **Medical advice or diagnosis** - This dataset is for research purposes only
- **Legal advice** regarding cannabis regulations or compliance
- **Commercial product claims** without proper validation and testing
- **Direct medical decision-making** without healthcare professional oversight

## Dataset Structure

The dataset is organized in JSON batch files, each containing Q&A pairs with associated metadata:

```json
{
"paper_name": {
"chunk_0": {
"generated": [
{
"question": "What is the primary psychoactive compound in cannabis?",
"answer": "Δ9-tetrahydrocannabinol (THC) is the primary psychoactive compound..."
}
],
"context": "Source text chunk from research paper...",
"source_pdf": "cannabis_pharmacology_2023.pdf"
}
}
}
```

**File Organization:**
- Batch files: `science_training_batch_001.json` through `science_training_batch_XXX.json`
- Each batch contains 5 processed documents
- Total file size: ~2.5GB across all batches

## Dataset Creation

### Curation Rationale

This dataset addresses a critical gap in domain-specific training data for cannabis science. Key motivations include:

- **Scientific accuracy**: Ensuring AI models have access to peer-reviewed cannabis research
- **Industry support**: Helping cannabis operators make informed, science-based decisions
- **Educational advancement**: Supporting research and education in cannabis science
- **Knowledge accessibility**: Making complex scientific literature more accessible through AI

### Source Data

- **400+ peer-reviewed research papers** from academic journals
- **College-level textbooks** on cannabis science and related fields
- **Academic publications** covering 2010-2024 research
- **Selection criteria**: Peer-reviewed, scientific rigor, relevance to cannabis research

#### Data Collection and Processing

1. **Document Processing**: Docling library for high-quality PDF conversion
2. **Intelligent Chunking**: HybridChunker for context-aware text segmentation
3. **Contextualization**: Each chunk enriched with surrounding document context
4. **Q&A Generation**: GPT-4o-mini with specialized prompts (5 pairs per chunk)
5. **Quality Control**: Structured JSON validation and rate-limited processing
6. **Parallel Processing**: 30 workers with 490 RPM rate limiting

#### Who are the source data producers?

- **Academic researchers** from universities and research institutions
- **Peer-reviewed journal publishers** in chemistry, biology, and medical fields
- **Scientific community members** specializing in cannabis research
- **Educational institutions** producing cannabis science curricula

## Bias, Risks, and Limitations

### Potential Biases
- **Academic bias**: Reflects published research perspectives and methodologies
- **Geographic bias**: Primarily Western/English-language research sources
- **Temporal bias**: Weighted toward more recent research (2015-2024)
- **Research focus bias**: May emphasize certain cannabis applications over others

### Risks and Limitations
- **Generated content accuracy**: AI-generated Q&A pairs may contain factual errors
- **Medical applications**: Not suitable for direct medical decision-making
- **Regulatory compliance**: Does not provide legal or regulatory guidance
- **Technical limitations**: Context window constraints during generation process
- **Model hallucinations**: Potential for GPT model to generate plausible but incorrect information
- **Coverage gaps**: Some specialized subtopics may be underrepresented

### Recommendations

Users should:
- Verify critical information against original sources
- Use for research and educational purposes only
- Consult healthcare professionals for medical applications
- Fact-check generated content for high-stakes applications

## Citation

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

```bibtex
@dataset{finney2025cannabis,
title={Cannabis Science Literature QA Dataset: 161K Question-Answer Pairs from Peer-Reviewed Research},
author={Kellan Finney},
year={2025},
publisher={Hugging Face},
url={https://huggingface.co/datasets/KellanF89/Cannabis_Science_Data}
}
```

## Dataset Card Authors

**Kellan Finney** - Dataset creation, curation, and processing pipeline development

## Dataset Card Contact

For questions, collaborations, or feedback, please reach out via [LinkedIn](https://www.linkedin.com/in/kellan-finney-m-s-861379a1).

---

*This dataset represents a significant advancement in making cannabis science knowledge accessible to AI systems, supporting both research progress and practical applications in the evolving cannabis industry.*