BIOMAT-NER Corpus: Train and Validation Sets

BIOMAT-NER Train and Validation Sets

BIOMAT-NER is a corpus developed within the scope of the Horizon Europe BIOMATDB project to support the extraction and classification of biomaterials-related concepts from the scientific literature. It focuses on the annotation of chemical substances, compounds, and material types—including trade names—relevant to the field of biomaterials. The corpus was created through a collaborative effort involving domain experts, who were tasked with the establishment of comprehensive and accurate annotation guidelines for the manual annotation of the final gold standard corpus. On this basis, PubMed abstracts were carefully selected based on relevant MeSH (Medical Subject Headings) categories associated with biomaterials and related disciplines to reflect the terminology commonly used in biomaterials research and manually annotated according to the predefined rules in the annotation guidelines.

This repository contains the train (4,553 documents) and validation (607 documents) sets of the BIOMAT-NER Corpus, which are made available under open access for public use. These sets have been released to support the development of Named Entity Recognition (NER) models for biomaterials-related concept extraction from scientific literature, particularly those mentions referring to chemical substances and compounds, and material types.

The test set is not included in this repository, as it is reserved for a future shared task planned within the scope of the project. For this reason, access to the full corpus remains restricted, but will be made publicly available upon completion of the shared task.

Resources

Project Website

Biomaterials Marketplace

Biomaterials Database

BIOMAT-NER full corpus

Annotation Guidelines