An Ensemble Method for Spelling Correction in Consumer Health Questions

Area of Science:

• Natural Language Processing
• Health Informatics
• Computational Linguistics

Background:

• Informal texts, like consumer health questions, often contain orthographic and grammatical errors.
• These errors hinder the automatic interpretation of crucial health information.
• Existing systems like ESpell struggle with the unique challenges of consumer health queries.

Purpose of the Study:

• To develop and evaluate a novel method for detecting and correcting orthographic errors in consumer health questions.
• To improve the accuracy of automatic interpretation systems for health-related queries.
• To assess the contribution of different features (orthographic, frequency, contextual) in error correction.

Main Methods:

• A hybrid approach combining edit distance and frequency counts with contextual similarity.
• Detection and correction of misspellings, word breaks, and punctuation errors.
• Evaluation on a dataset of spell-corrected consumer health questions from the NLM collection.

Main Results:

• The proposed method achieved a F1 score of 0.61.
• This significantly outperforms the informed baseline of 0.29 achieved by ESpell.
• Orthographic similarity was identified as the most critical feature for correction.

Conclusions:

• The developed method effectively corrects orthographic errors in consumer health questions.
• Orthographic similarity, frequency, and contextual information collectively enhance correction accuracy.
• This work advances the automatic processing of user-generated health information.