Similarity-based-Unsupervised-Spelling-Correction-Using-BioWordVec

This is similarity-based unsupervised spelling correction using BioWordVec for bacteria culture and antimicrobial susceptibility reports.
Please visit the original repo of BioWordVec (Yijia Zhang, et al.) for more information about the pre-trained BioWordVec.

Environments

python 3
numpy 1.18
pandas 1.1
keras 2.4.3
gensim 3.8.3
pyxDamerauLevenshtein 1.6.1
symspellpy 6.5

Pretrained Embedding

• BioWordVec
  https://ftp.ncbi.nlm.nih.gov/pub/lu/Suppl/BioSentVec/BioWordVec_PubMed_MIMICIII_d200.vec.bin

Usage

Example:

python SUSC_run.py --data sample.csv

Arguments:

--data				Data 
--similar_n			Most similar words 
--fasttext_min_similarity	Cosine similarity 
--edit_distance_threshold	Edit distance 

Overall Architecture

Citation

If you find this work useful for your research, please cite our paper:

Kim T, Han S, Kang M, Lee S, Kim J, Joo H, Sohn J
Similarity-Based Unsupervised Spelling Correction Using BioWordVec: Development and Usability Study of Bacterial Culture and Antimicrobial Susceptibility Reports
JMIR Med Inform 2021;9(2):e25530
URL: https://medinform.jmir.org/2021/2/e25530
DOI: 10.2196/25530