Similarity-Based Unsupervised Spelling Correction Using BioWordVec: Development and Usability Study of Bacterial Culture and Antimicrobial Susceptibility Reports

Abstract

Background Existing bacterial culture test results for infectious diseases are written in unrefined text, resulting in many problems, including typographical errors and stop words. Effective spelling correction processes are needed to ensure the accuracy and reliability of data for the study of infectious diseases, including medical terminology extraction. If a dictionary is established, spelling algorithms using edit distance are efficient. However, in the absence of a dictionary, traditional spelling correction algorithms that utilize only edit distances have limitations. Objective In this research, we proposed a similarity-based spelling correction algorithm using pretrained word embedding with the BioWordVec technique. This method uses a character-level N-grams–based distributed representation through unsupervised learning rather than the existing rule-based method. In other words, we propose a framework that detects and corrects typographical errors when a dictionary is not in place. Methods For detected typographical errors not mapped to Systematized Nomenclature of Medicine (SNOMED) clinical terms, a correction candidate group with high similarity considering the edit distance was generated using pretrained word embedding from the clinical database. From the embedding matrix in which the vocabulary is arranged in descending order according to frequency, a grid search was used to search for candidate groups of similar words. Thereafter, the correction candidate words were ranked in consideration of the frequency of the words, and the typographical errors were finally corrected according to the ranking. Results Bacterial identification words were extracted from 27,544 bacterial culture and antimicrobial susceptibility reports, and 16 types of spelling errors and 914 misspelled words were found. The similarity-based spelling correction algorithm using BioWordVec proposed in this research corrected 12 types of typographical errors and showed very high performance in correcting 97.48% (based on F1 score) of all spelling errors. Conclusions This tool corrected spelling errors effectively in the absence of a dictionary based on bacterial identification words in bacterial culture and antimicrobial susceptibility reports. This method will help build a high-quality refined database of vast text data for electronic health records.

Year 2021: COVID-19, Information Extraction and BERTization among the Hottest Topics in Medical Natural Language Processing

“…To overcome these limitations, several alternative solutions have been addressed this year by researchers: fine-tuning of existing models [58][59][60][61][62][63], domain adaptation [43,64], transfer learning [48,50,60], self-training [43]. To go further in these directions, new trends were also observed in 2021, like reuse of older architectures based on fastText [65] and word2vec [66] enriched with basic language information: orthographic and lexical [67], syntactic-semantic classes [68], medical knowledge [46], subword embeddings [14,69], vector retrofitting [67,68]. Note also that multi-task systems were also proposed and can satisfy several NLP tasks adapted to the medical area [41].…”