The second set includes n-grams created using dummy words to replace certain
concepts in the sentences. For example, “In 1970 furosemide was reported to cause. . .”
would be replaced by “In was reported to cause. . . .” The authors tested
several machine-learning algorithms including ZeroR, Naive Bayes, Decision Tree, and
Support Vector Machines (SVM) using a combination of features and found that the
best performance was obtained using SVM and a combination of statistical features
and n-grams up to 3.