Update README.md

README.md

@@ -17,15 +17,5 @@ Using the UMLS and the SNOMED-CT ontology we identified *hard* positive term pai
 and *soft* positive term pairs by
 * using the known positive relationships between CUIs
 
-We expand this with negative term pairs by random sampling, followed by selecting the top-N most dissimilar terms with (sim<0.25)
-* using [BioLORD 2023 M Dutch](https://huggingface.co/FremyCompany/BioLORD-2023-M-Dutch-InContext-v1)
-* using [sapMedRoBERTa.nl](https://huggingface.co/UMCU/sap_umls_medroberta.nl_meantoken)
-* using [RotatE graph embedding](https://arxiv.org/abs/1902.10197)
-
-We further expand these lists using LLMs.
-
-We now end up with.
-* ontology-based same-concept positive pairs, extended with LLMs
-* ontology-based related-concept positive pairs, extended with LLMs
-* BioLoRD/RotatE/sap-based negatives, extended with LLMs
+As future work we want to expand the positives using e.g. paraphrasing and context selection from existing clinical documents (like KRISSBERT).