Abstract

While state of the art techniques can address the problem of automatically detecting negated medical observations, negation using the word `not' presents a harder problem than other kinds of negation. We apply machine learning techniques to distinguish sentences where the word `not' does and does not negate a medical observation. Our corpus contains hospital reports such as progress notes and emergency room notes. We use two different machine learning algorithms, Naive Bayes and Decision Trees, and both achieve significant improvement over the baseline. We also analyze the data and the classifiers' behavior and output to learn more about the problem and the usefulness of various features in our feature vector.

Citations