Figure 6

Distance between tokens appearing in private and public medical texts. Vector space model was used to capture the similarities (distance) in vector space between tokens in the private training set and tokens in the public publications set. In total, 669 physician notes and 669 medical publications were analyzed for their pairwise distances. An equal number of public (non-PHI) tokens and private tokens were selected from train. The boxplot shows 25th and 75th percentiles for distances from publication tokens to training tokens. The leftmost column reveals that public terms that are not PHI are more similar to publication tokens than any other group.