B

Biomednlp KRISSBERT PubMed UMLS EL

Developed by microsoft
KRISSBERT is a knowledge-enhanced self-supervised learning model for biomedical entity linking. It trains contextual encoders using unannotated text and domain knowledge to effectively address the diversity and ambiguity of entity names.
Downloads 4,643
Release Time : 4/15/2022

Model Overview

KRISSBERT is a biomedical entity linking model that understands context and accurately links to standardized entity IDs (such as CUIs in UMLS), solving problems traditional methods face with unseen entities and lack of contextual understanding.

Model Features

Knowledge-enhanced Self-supervised Learning
Utilizes UMLS ontology biomedical entity names and PubMed abstracts for self-supervised pre-training without requiring gold-standard entity mention examples or canonical descriptions of all entities.
Contextual Understanding
Capable of understanding the context of entity mentions to accurately disambiguate and link to standardized entity IDs, rather than merely predicting surface forms.
High Performance
Achieves state-of-the-art performance on seven standard biomedical entity linking datasets, with accuracy improvements of up to 20 percentage points over previous self-supervised methods.

Model Capabilities

Biomedical Entity Linking
Contextual Understanding
Entity Disambiguation

Use Cases

Biomedical Research
Medical Literature Entity Linking
Links entity mentions in medical literature to standardized entity IDs in UMLS, such as linking "ER" to "Emergency Room" or "Estrogen Receptor Gene" based on context.
Achieves approximately 58.3% Top-1 accuracy on the MedMentions dataset.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase