Figure 7.

Confusion matrices for both alprazolam and fentanyl queries. True labels are determined by manual labeling. Predicted labels are determined by (a) classifying all generated terms as true; (b) classifying all generated terms that appear in RedMed as true; (c) classifying all generated terms that pass the drug name filter as true; (d) classifying all generated terms that pass the drug name filter and the generation frequency filter as true; (e) classifying all generated terms that pass the drug name filter and the Google search filter as true; and (f) classifying all generated terms that pass the drug name filter, the generation frequency filter, and the Google search filter as true. In each confusion matrix, a 0 denotes a negative classification, which is a non-synonym, and a 1 denotes a positive classification, which is a synonym.