Jaymerry/itis-taxonomy-instruct-30k

ITIS Taxonomy Instruction Dataset (30K)

Overview

The ITIS Taxonomy Instruction Dataset (30K) is a structured instruction-response dataset derived from the public domain Integrated Taxonomic Information System (ITIS) database.

It contains 30,000 instruction-output pairs designed for fine-tuning large language models on taxonomic reasoning and biodiversity-related tasks.

The dataset supports offline taxonomic intelligence use cases, including classification retrieval, lineage reconstruction, rank identification, and synonym validation.

---

Data Source

This dataset is derived from:

**Integrated Taxonomic Information System (ITIS)**
https://www.itis.gov/

ITIS is a public domain database maintained by U.S. federal agencies and partners.

All taxonomic data used in this dataset are sourced strictly from ITIS structured records.

---

License

This dataset is released under:

CC0 1.0 Universal (Public Domain Dedication)

The original ITIS data are public domain.
This dataset is a structured transformation of those public domain records.

---

Dataset Structure

Each record follows an Alpaca-style instruction format:

{
  "instruction": "What taxonomic rank is Panthera leo?",
  "input": "",
  "output": "Species"
}

For lineage tasks, the output may contain structured JSON:

{
  "instruction": "Provide the full taxonomic classification (lineage) of Panthera leo.",
  "input": "",
  "output": "{"tsn":12345,"scientific_name":"Panthera leo","rank":"Species","status":"valid","lineage":[...]}"
}

---

Splits

The dataset contains:

• train.jsonl
• validation.jsonl
• test.jsonl

Each file is in JSONL format (one JSON object per line).

---

Supported Task Types

The dataset includes multiple taxonomic reasoning tasks:

1. Rank Identification

• "What taxonomic rank is X?"

2. Lineage Reconstruction

• "Provide the full taxonomic classification of X."

3. Validity Check

• "Is X a valid/accepted taxon in ITIS?"

4. Parent Taxon Identification

• "Which immediate parent taxon does X belong to?"

5. Common Name ↔ Scientific Name Mapping

• "What is the scientific name of the common name 'lion'?"

6. Taxonomic Comparison

• "Do X and Y belong to the same genus?"

---

Intended Use

This dataset is intended for:

• Fine-tuning large language models (LLMs)
• Offline biodiversity assistants
• Scientific taxonomy QA systems
• Knowledge graph construction
• Educational tools in biological classification

---

Out-of-Scope Use

This dataset does not contain:

• Ecological descriptions
• Conservation status
• Species distribution data
• Morphological data
• Narrative biological text

It is strictly focused on structured taxonomic hierarchy.

---

Data Generation Process

The dataset was generated by:

1. Parsing the ITIS SQLite database
2. Extracting taxonomic units, ranks, parent relationships, and vernacular names
3. Reconstructing lineage hierarchies
4. Programmatically generating instruction-response pairs
5. Filtering empty or incomplete outputs
6. Splitting into train/validation/test sets

No external knowledge sources (e.g., Wikipedia) were used.

---

Known Limitations

• English vernacular names only
• Some rare taxonomic edge cases may not include vernacular names
• Outputs are derived strictly from ITIS structure and may not reflect the latest taxonomic revisions outside ITIS

---

Citation

If you use this dataset, please cite:

Integrated Taxonomic Information System (ITIS)
https://www.itis.gov/

---

Maintainer

Dataset prepared and structured by Jay Merry.