Table 4.
The performance of CRF-based and SSVM-based CEM systems when different types of WR features were used.
| Feature | CRF | SSVM | ||||
|---|---|---|---|---|---|---|
| Precision | Recall | F-measure | Precision | Recall | F-measure | |
| Baseline | 89.91 | 79.77 | 84.54 | 88.95 | 81.32 | 84.96 |
| Baseline + BC | 90.01 | 80.50 | 84.99 | 89.04 | 81.51 | 85.11 |
| Baseline + RI | 89.52 | 80.60 | 84.83 | 88.43 | 82.03 | 85.12 |
| Baseline + SKIP | 89.86 | 80.71 | 85.04 | 88.88 | 81.73 | 85.15 |
| All | 89.42 | 81.08 | 85.05 | 88.34 | 82.27 | 85.20 |
* BC, RI and SKIP denote three different types of word representation features: Brown clustering-based, random indexing-based and skip-gram-based word representations respectively.