metadata
dataset_info:
features:
- name: messages
list:
- name: content
dtype: string
- name: role
dtype: string
- name: source
dtype: string
- name: ref_count
dtype: int64
- name: languages
list: string
splits:
- name: train_single
num_bytes: 37115180
num_examples: 25772
- name: valid_single
num_bytes: 1861539
num_examples: 1235
- name: train_group
num_bytes: 12511678
num_examples: 725
- name: valid_group
num_bytes: 987781
num_examples: 31
download_size: 11510722
dataset_size: 52476178
configs:
- config_name: default
data_files:
- split: train_single
path: data/train_single-*
- split: valid_single
path: data/valid_single-*
- split: train_group
path: data/train_group-*
- split: valid_group
path: data/valid_group-*
Reference Parsing Dataset for LoRA Training
This dataset contains structured reference parsing examples for fine-tuning language models with LoRA.
Dataset Description
Two complementary datasets for training reference parsing models:
- Single: Parse individual bibliographic references (25,772 train / 1,235 valid)
- Group: Extract and parse references from documents (725 train / 31 valid)
Data Sources
- LinkedBook: 24,615 train / 1,055 valid tagged references (multilingual: Italian, English, French, German, Spanish)
- CEX: 16 academic papers from 27 categories (96 reserved for testing)
- EXCITE: 35 papers from 3 reference location classes (316 reserved for testing)
Format
Conversation-style JSON with system, user, and assistant messages. The assistant outputs structured JSON with reference fields including authors, title, journal, volume, pages, date, publisher, etc.