BioASQ Participants Area
BioNNE-R Shared Task at BioASQ 2026
Shared Task Overview
The BioNNE-R Shared Task addresses a challenging and underexplored problem in biomedical Natural Language Processing (NLP): relation extraction involving nested named entities, where entity mentions may overlap or contain other entity mentions within their boundaries. Such structures frequently occur in biomedical texts and complicate traditional relation extraction approaches.
Participants are required to develop models for nested relation extraction in English, Russian, or in a bilingual setting, depending on the selected track.
Goal: The objective of the task is to extract predefined semantic relations between annotated nested biomedical entity mentions. The problem is formulated as a multi-class relation classification task over entity pairs.
Data: The dataset consists of English and Russian scientific abstracts in the biomedical domain, annotated with nested entities and semantic relations between them. The BioNNE-R task utilizes the entity annotation of the NEREL-BIO dataset, which includes annotated mentions of disorders, anatomical structures, chemicals, diagnostic procedures, and related biomedical concepts. The dataset is specifically designed to reflect the complex structure of nested entity mentions and the partial, compositional nature of medical terminology.
Participants are allowed to use any model architecture and any publicly available resources or datasets to achieve the best possible performance.
Evaluation Tracks
The task consists of three subtasks grouped into two evaluation tracks:
-
Monolingual Track:
- Subtask 1 - English
- Subtask 2 - Russian
- Bilingual Track (Subtask 3): A single multilingual model must be trained on the combined English and Russian datasets.
Systems are evaluated as a multi-class classification problem using macro-averaged F1-score over all relation types.
To participate in BioNNE-R, please register via the BioASQ registration page: BioASQ Registration .
The task is open from February 2026 onwards. The competition is hosted on Codabench: Codabench Competition Page .
Detailed information about the dataset structure, submission format, evaluation protocol, and important dates is available on the official GitHub repository: BioNNE-R GitHub Page .
