Table 3.
Features | Precision | Recall | F1 |
---|---|---|---|
Baseline | 82.1 | 71.4 | 76.4 |
+ Character features | 75.6 | 74.6 | 77.9 |
+ Drug features | 83.1 | 74.0 | 78.3 |
+ EHR embedding clusters (extended) | 82.6 | 75.2 | 78.7 |
+ NoEHR embedding clusters (extended) | 82.1 | 75.6 | 78.7 |
+ EHR and NoEHR embedding clusters (extended) | 82.6 | 76.4 | 79.3 |
+ All features (standard) | 82.8 | 76.7 | 79.6 |
+ All features (extended) | 83.8 | 78.1 | 80.9 |
Baseline features were comprised of commonly used NER features such as tokens, stems, parts of speech and lexical patterns of capitalization, digits, and punctuation
EHR electronic health record, NER named entity recognition