shengz commited on
Commit
cfd4687
1 Parent(s): d889b40

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +2 -2
README.md CHANGED
@@ -12,10 +12,10 @@ widget:
12
 
13
  Entity linking faces significant challenges such as prolific variations and prevalent ambiguities, especially in high-value domains with myriad entities. Standard classification approaches suffer from the annotation bottleneck and cannot effectively handle unseen entities. Zero-shot entity linking has emerged as a promising direction for generalizing to new entities, but it still requires example gold entity mentions during training and canonical descriptions for all entities, both of which are rarely available outside of Wikipedia ([Logeswaran et al., 2019](https://aclanthology.org/P19-1335.pdf); [Wu et al., 2020](https://aclanthology.org/2020.emnlp-main.519.pdf)). We explore Knowledge-RIch Self-Supervision (KRISS) and train a contextual encoder (KRISSBERT) for entity linking, by leveraging readily available unlabeled text and domain knowledge.
14
 
15
- This KRISSBERT is initialized with [PubMedBERT](https://huggingface.co/microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract) parameters, and then trained using self-supervised examples that are generated by combining [PubMed](https://pubmed.ncbi.nlm.nih.gov/) abstracts and the [UMLS](https://www.nlm.nih.gov/research/umls/index.html) ontology. Experiments on seven standard biomedical entity linking datasets show that KRISSBERT attains new state of the art, outperforming prior self-supervised methods by as much as 20 absolute points in accuracy.
16
  See [Zhang et al., 2021](https://arxiv.org/abs/2112.07887) for the details.
17
 
18
- Note that some prior work like [BioSyn](https://aclanthology.org/2020.acl-main.335.pdf), [SapBERT](https://aclanthology.org/2021.naacl-main.334.pdf), and their follow-up ([Lai et al., 2021](https://aclanthology.org/2021.findings-emnlp.140.pdf)) claimed to do entity linking, but their systems completely ignore the context of an entity mention, and can only predict a surface form, _**not CUI**_ (See Figure 1 in [BioSyn](https://aclanthology.org/2020.acl-main.335.pdf)). Therefore, they can't disambiguate ambiguous mentions. For instance, given the entity mention "_ER_" in the sentence "*ER crowding has become a wide-spread problem*", their systems predict the nearest entity name (which is also "ER") in the ontology. They can't pinpoint the target entity "*Emergency Room (C0562508)*", because other entities such as "*Estrogen Receptor Gene (C1414461)*" and "*Endoplasmic Reticulum(C0014239)*" also use "ER" as their alias. Without using the context information, their systems can't resolve such ambiguity. Unfortunately, their evaluation considers it a correct prediction, for the reason that "ER" matches one of the aliases of the gold entity, which is problematic. Consequently, the reported results in their papers do not reflect true performance on entity linking.
19
 
20
  ## Citation
21
 
 
12
 
13
  Entity linking faces significant challenges such as prolific variations and prevalent ambiguities, especially in high-value domains with myriad entities. Standard classification approaches suffer from the annotation bottleneck and cannot effectively handle unseen entities. Zero-shot entity linking has emerged as a promising direction for generalizing to new entities, but it still requires example gold entity mentions during training and canonical descriptions for all entities, both of which are rarely available outside of Wikipedia ([Logeswaran et al., 2019](https://aclanthology.org/P19-1335.pdf); [Wu et al., 2020](https://aclanthology.org/2020.emnlp-main.519.pdf)). We explore Knowledge-RIch Self-Supervision (KRISS) and train a contextual encoder (KRISSBERT) for entity linking, by leveraging readily available unlabeled text and domain knowledge.
14
 
15
+ Specifically, the KRISSBERT model is initialized with [PubMedBERT](https://huggingface.co/microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract) parameters, and then continuously pretrained using biomedical entity names from the [UMLS](https://www.nlm.nih.gov/research/umls/index.html) ontology to self-supervise entity linking examples from [PubMed](https://pubmed.ncbi.nlm.nih.gov/) abstracts. Experiments on seven standard biomedical entity linking datasets show that KRISSBERT attains new state of the art, outperforming prior self-supervised methods by as much as 20 absolute points in accuracy.
16
  See [Zhang et al., 2021](https://arxiv.org/abs/2112.07887) for the details.
17
 
18
+ Note that some prior systems like [BioSyn](https://aclanthology.org/2020.acl-main.335.pdf), [SapBERT](https://aclanthology.org/2021.naacl-main.334.pdf), and their follow-up work (e.g., [Lai et al., 2021](https://aclanthology.org/2021.findings-emnlp.140.pdf)) claimed to do entity linking, but their systems completely ignore the context of an entity mention, and can only predict a surface form in the entity dictionary (See Figure 1 in [BioSyn](https://aclanthology.org/2020.acl-main.335.pdf)), _**not the canonical entity ID (e.g., CUI in UMLS)**_. Therefore, they can't disambiguate ambiguous mentions. For instance, given the entity mention "_ER_" in the sentence "*ER crowding has become a wide-spread problem*", their systems ignore the sentence context, and simply predict the closest surface form, which is just "ER". Multiple entities share this surface form as a potential name or alias, such as *Emergency Room (C0562508)*, *Estrogen Receptor Gene (C1414461)*, and *Endoplasmic Reticulum(C0014239)*. Without using the context information, their systems can't resolve such ambiguity and pinpoint the correct entity *Emergency Room (C0562508)*. More problematically, their evaluation would deem such an ambiguous prediction as correct. Consequently, the reported results in their papers do not reflect true performance on entity linking.
19
 
20
  ## Citation
21