Due to differences on dataset and language, we design positive and negative sets for each dataset with slightly different heuristics.
Run all blocks of Contrastive_sets_construction_HQS.ipynb
to get the contrastive sets of HQS.
Follow these two steps:
python Contrastive_sets_construction_RRS.py
python ad-hoc_cleanup_RRS.py
For performance reason, we use hoc_cleanup_RRS.py
to clean up the newly generated positive sets from Contrastive_sets_construction_RRS.py
.
Run all blocks of Contrastive_sets_construction_MDS.ipynb
to get the contrastive sets of MDS. Note that you need ALL_medical_term_file_Chinese.txt
to build the contrastive set. It is a file that contains all the medical terms from training, validation, and test splits of MDS (the collection of 1_medical
and 2_medical
). We share it here to give everyone an idea how it looks like.