Abbreviation disambiguation using semantic abstraction of symbols and numeric terms

We propose an abbreviation disambiguation approach that utilizes semantic representation of symbols and numeric terms as well as the words in clinical documents. While majority of related works treats symbols and numeric words as stopword, we show that they play an important role especially in coarse-grained documents such as CDA (clinical document architecture) documents, which contain a lot of jargons, symbols, and abbreviations written by doctors. For abbreviation disambiguation task using a classifier, we compared several variations of our approach with a traditional bag-of-words method. The results show that the system using semantic abstraction of symbols and numeric terms can improve the accuracy from 87.9% to 92.6% when a SVM classifier is used.