Automated Spelling Correction for Clinical Text Mining in Russian.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Medical Informatics

Background:

• Clinical text presents unique challenges for spell checking due to specialized vocabulary and potential for errors.
• Accurate spell checking is crucial for reliable medical text mining and analysis.
• Existing tools may not adequately address the nuances of Russian clinical language.

Purpose of the Study:

• To develop and evaluate a novel spell checker module specifically designed for Russian clinical text.
• To integrate this spell checker into a broader medical text mining tool.
• To improve the accuracy of automated analysis of clinical documents.

Main Methods:

• Combination of string distance measure algorithms (e.g., Levenshtein, Damerau-Levenshtein) for initial error detection.
• Application of machine learning embedding methods (e.g., Word2Vec, FastText) to understand contextual relevance and word meaning.
• Development of a hybrid approach leveraging both algorithmic and machine learning techniques.

Main Results:

• Achieved an overall precision of 0.86 for the spell checker module.
• Demonstrated high lexical precision of 0.975, indicating accurate identification of correct terms.
• Reported an error precision of 0.74, showing effectiveness in identifying and correcting misspellings.
• The spell checker is a component of a larger medical text mining tool.

Conclusions:

• The developed spell checker module significantly enhances the accuracy of processing Russian clinical text.
• The hybrid approach combining string distance and machine learning is effective for clinical spell correction.
• This advancement supports more reliable medical text mining, including detection of misspelling, negation, experiencer, and temporality.