|
|
--- |
|
|
dataset_info: |
|
|
features: |
|
|
- name: title |
|
|
dtype: string |
|
|
- name: uuid |
|
|
dtype: string |
|
|
- name: pmc_id |
|
|
dtype: string |
|
|
- name: search_term |
|
|
dtype: string |
|
|
- name: text |
|
|
dtype: string |
|
|
splits: |
|
|
- name: train |
|
|
num_bytes: 1593278889 |
|
|
num_examples: 33673 |
|
|
- name: test |
|
|
num_bytes: 82702112 |
|
|
num_examples: 1773 |
|
|
download_size: 625424007 |
|
|
dataset_size: 1675981001 |
|
|
configs: |
|
|
- config_name: default |
|
|
data_files: |
|
|
- split: train |
|
|
path: data/train-* |
|
|
- split: test |
|
|
path: data/test-* |
|
|
--- |
|
|
|
|
|
This is the Panancea PMC Articles Dataset, a dataset consisting of medical literature sampled from PubMed Central, covering different medical specialties, used to train Panacea, our medical LLM finetune. |
|
|
|
|
|
|
|
|
# Dataset Size |
|
|
| Metric | Size | |
|
|
|----------------|-------------| |
|
|
| # of Entries | 35,446 | |
|
|
| # of Words | 250.6M | |
|
|
| Storage Size | 1.68 GB | |
|
|
|
|
|
|
|
|
# Dataset Composition |
|
|
Our dataset is composed of articles from these specialties. We query them from PubMed Central using Medical Subject Headings (MeSH), a controlled vocabulary thesaurus for expanding article searches in PubMed Central. |
|
|
| Category | Count | Percentage (%) | |
|
|
|--------------------------------|-------|----------------| |
|
|
| Anatomy | 4555 | 12.85 | |
|
|
| Surgery | 4381 | 12.36 | |
|
|
| Family Medicine | 3853 | 10.87 | |
|
|
| Medicine | 3748 | 10.57 | |
|
|
| Microbiology | 2806 | 7.92 | |
|
|
| Pathology | 2397 | 6.76 | |
|
|
| Dentistry | 2261 | 6.38 | |
|
|
| Biochemistry | 2010 | 5.67 | |
|
|
| Pediatrics | 1628 | 4.59 | |
|
|
| Pharmacology | 1601 | 4.52 | |
|
|
| Physiology | 1273 | 3.59 | |
|
|
| Ophthalmology | 1156 | 3.26 | |
|
|
| Social and Preventive Medicine| 1030 | 2.91 | |
|
|
| Forensic Medicine | 1023 | 2.89 | |
|
|
| Gynaecology | 812 | 2.29 | |
|
|
| ENT | 490 | 1.38 | |
|
|
| Psychiatry | 419 | 1.18 | |
|
|
| Radiology | 3 | 0.01 | |
|
|
| **Total** | 35446 | 100 | |
|
|
|
|
|
|
|
|
# Dataset Preprocessing |
|
|
- We have only briefly cleaned the datasets, by excluding figures, captions, references, front and end materials from the returned XML file, storing the results as .txt files. |
|
|
- Some Regex filtering is done to take away special characters and Latex symbols. |
