BioConsistency

Full description of the adversarial data generation is illustrated in paper:

Directories for data repositories

The BC4GO corpus are stored within /corpus/annotation/.xml and /corpus/annotation/.txt
Consistent instances are stored within /corpus/positives/*.txt
Each *.txt file corresponds to an individual article in BC4GO named by it's PMID

(A) Contradictory inconsistencies are stored within /corpus/go_swapped_negatives/*.txt
(B) Over-specific inconsistencies are stored within /corpus/go_replaced_negatives/*.txt
(C) Unsupportive inconsistencies are stored within /corpus/sent2vec_pairs/*.txt. (The first line of each record is a consistent instance extracted from BC4GO. Following by lines of evidence text replaced instances as Unsupportive inconsistencies. The first attribute are the cosine similarity score between BioSentVec of evidence text and BioSentVec of unsupportive text)

(D) Error-code inconsistencies are stored within /corpus/ec_replaced_netatives/*.txt

Change the directory for data depository in run.sh file
Find select_go_child() function in /adversarial/tweak_goa.py and comment lines as instructed to able/disable application of training set optimisation
run run.sh to re-generate instances if needed

Name		Name	Last commit message	Last commit date
Latest commit History 23 Commits
adversarial		adversarial
corpus		corpus
obj		obj
similarity		similarity
spliter		spliter
tool		tool
.DS_Store		.DS_Store
.gitignore		.gitignore
README.md		README.md
__init__.py		__init__.py
run.sh		run.sh