This repository contains the code for the creation and evaluation of two domain-specific Dutch Medical Language models. Most data used in this project can not be provided due to privacy constraints. Where possible, data is provided.
The src folder contains all code and data. Per subfolder, a readme is provided. The subfolders it contains are the following:
└───src
│ └───gather_traindata (provides the code used for gathering, filtering and preparing the data used for pre-training in train_lm)
│ └───train_lm (provides the code to pre-train two medical language models: from scratch and extending RobBERT
│ └───ICF_test (provides the code to fine-tune and test language models on a medical classification task)
│ └───similarity_test (provides the code to create a similarity test set from hospital notes and provides the code and data to test language models on this)
│ └───NER_test (provides the code to fine-tune and test language models on named entitiy recognition for general Dutch)
│ └───anonymization (provides the code to anonymize the vocabulary of a language model and test the level of anonymicity of a language model)
For more information or access to a copy of the full thesis you can contact me at [email protected]