SapBERT-DE is a model for German biomedical entity linking which is obtained by fine-tuning multilingual entity linking model cambridgeltl/SapBERT-UMLS-2020AB-all-lang-from-XLMR using a German biomedical entity linking knowledge base named UMLS-Wikidata.
Usage
import numpy as np
from tqdm import tqdm
import torch
from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("permediq/SapBERT-DE", use_fast=True)
model = AutoModel.from_pretrained("permediq/SapBERT-DE").cuda()
# entity descriptions to embed
entity_descriptions = ["Cerebellum", "Zerebellum", "Kleinhirn", "Anaesthesie"]
bs = 32 # batch size
all_embs = []
for i in tqdm(np.arange(0, len(entity_descriptions), bs)):
toks = tokenizer.batch_encode_plus(entity_descriptions[i:i+bs],
padding="max_length",
max_length=40, # model trained with 40 max_length
truncation=True,
return_tensors="pt")
toks_cuda = {}
for k,v in toks.items():
toks_cuda[k] = v.cuda()
cls_rep = model(**toks_cuda)[0][:,0,:]
all_embs.append(cls_rep.cpu().detach())
all_embs = torch.cat(all_embs)
def cos_sim(a, b):
a_norm = torch.nn.functional.normalize(a, p=2, dim=1)
b_norm = torch.nn.functional.normalize(b, p=2, dim=1)
return torch.mm(a_norm, b_norm.transpose(0, 1))
# cosine similarity of first entity with all the entities
print(cos_sim(all_embs[0].unsqueeze(0), all_embs))
# >>> tensor([[1.0000, 0.9337, 0.6206, 0.2086]])
BibTeX
@inproceedings{mustafa-etal-2024-leveraging,
title = "Leveraging {W}ikidata for Biomedical Entity Linking in a Low-Resource Setting: A Case Study for {G}erman",
author = "Mustafa, Faizan E and
Dima, Corina and
Ochoa, Juan and
Staab, Steffen",
booktitle = "Proceedings of the 6th Clinical Natural Language Processing Workshop",
month = jun,
year = "2024",
address = "Mexico City, Mexico",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.clinicalnlp-1.17",
pages = "202--207",