Table 5.
Evaluation of SNoW and Stat De-id on recognizing people, locations, and organizations found in the random corpus. Note that these are the only entity types SNoW was built to recognize. Highest F-measures are in bold. The difference in PHI F-measures between SNoW and Stat De-id is not significant at α = 0.05. The difference in non-PHI F-measures is significant at the same α and marked as such with an *.
| Method | Class | Precision | Recall | F-measure |
|---|---|---|---|---|
| Stat De-id | PHI | 98.31% | 96.62% | 97.46% |
| SNoW | PHI | 95.18% | 97.63% | 96.39%* |
| Stat De-id | Non-PHI | 99.64% | 99.82% | 99.73% |
| SNoW | Non-PHI | 99.75% | 99.48% | 99.61%* |