Search

Background: Medical text repositories not only constitute a significant amount of data but also represent an interesting scientific test bed for those willing to apply natural language processing to information retrieval. In order to improve retrieval performance of the Catalogue and Index of Health Resources in French (CISMeF) and its search tool Doc’CISMeF, we tested a new method to correct misspellings of the queries written by the users. Methods: In addition to exact phonetic term matching, we tested two approximate string comparators. The approximate comparators are the string distance metric of Stoilos and the Levenshtein edit distance. We also calculated the results of the two-combined algorithm to examine whether it improves misspelling correction of the queries. Results: At a threshold comparator score of 0.2, the normalized Levenshtein algorithm achieved the highest recall of 76% but the highest precision 94% is achieved by combining the distances of Levenshtein and Stoilos. Conclusion: Although the well-known good performance of the normalized edit distance of Levenshtein, we have demonstrated in this paper that its combination with the Stoilos algorithm improves the results for misspelling correction.