Recognition Performance over the Test Corpus of the Modified JNLPBA Corpus, N = 6,241
| System | Notes | Precision | Recall | F-Measure |
| ABNER | Without post-processing | 0.6491 | 0.7505 | 0.6961 |
| LingPipe | 36-gram without post-processing | 0.6079 | 0.7166 | 0.6577 |
| MEMM | 2nd-order without post-processing | 0.6668 | 0.7375 | 0.7004 |
| CRF | Configured for BioCreAtIvE | 0.7083 | 0.7702 | 0.7379 |
| BioTagger-GM | Combination of four systems | 0.7058 | 0.8247 | 0.7607 |
-
GM = gene mention; CRF = conditional random field.
-
The JNLPBA corpus is different from the BioCreAtIvE II GM corpus in several ways. The corpus and modules were adjusted for these differences, but the taggers were not tuned to the corpus.









