Automatic structuring and correction suggestion system for Hungarian clinical records
Borbála Siklósi, György Orosz, Attila Novák, Gábor Prószéky · Repository of the Academy's Library (Library of the Hungarian Academy of Sciences) · 2012
The first steps of processing clinical documents are structuring and normalization. In this paper we demonstrate how we compensate the lack of any structure in the raw data by transforming simple formatting features automatically to structural units. Then we developed an algorithm to separate running text from tabular and numerical data. Finally we generated correcting suggestions for word forms recognized to be incorrect. Some evaluation results are also provided for using the system as automatically correcting input texts by choosing the best possible suggestion from the generated list. Our method is based on the statistical characteristics of our Hungarian clinical data set and on the HUMor Hungarian morphological analyzer. The conclusions claim that our algorithm is not able to correct all mistakes by itself, but is a very powerful tool to help manually correcting Hungarian medical texts in order to produce a correct text corpus of such a domain.