Table 3.
Feature contributions for the similarity metric
| Features | A | P | R | F1 | ΔF1 |
|---|---|---|---|---|---|
| Sim (ch) | 0.963 | 0.879 | 0.895 | 0.887 | |
| Sim (wd) | 0.937 | 0.844 | 0.747 | 0.793 | |
| Sim (ch + wd) | 0.962 | 0.877 | 0.890 | 0.884 | |
| Levenshtein | 0.939 | 0.849 | 0.754 | 0.799 | |
| Jaro–Winkler | 0.918 | 0.920 | 0.534 | 0.676 | |
| SoftTFIDF | 0.921 | 0.817 | 0.656 | 0.728 | |
| Full | 0.965 | 0.883 | 0.900 | 0.892 | |
| - Sim (ch) | 0.947 | 0.855 | 0.808 | 0.831 | −0.061 |
| - Sim (wd) | 0.965 | 0.885 | 0.898 | 0.891 | −0.001 |
| - Sim (ch+wd) | 0.950 | 0.868 | 0.810 | 0.838 | −0.054 |
| - Levenshtein | 0.965 | 0.882 | 0.898 | 0.890 | −0.002 |
| - Jaro–Winkler | 0.965 | 0.882 | 0.899 | 0.891 | −0.001 |
| - SoftTFIDF | 0.965 | 0.882 | 0.901 | 0.892 | −0.000 |
ΔF1, the difference of F1 score from the Full feature set; Sim (ch), character n-gram similarity; Sim (wd), word n-gram similarity