Table 2.
Performance for PK DDI extraction on the in vitro dataset
| Methods | P | R | F 1 |
|---|---|---|---|
| DEP | 43.43 % | 63.24 % | 51.50 % |
| PASa | 73.03 % | 62.07 % | 67.68 % |
| DEP_SCa | 70.32 % | 61.93 % | 65.86 % |
| PAS_SCa,b | 69.23 % | 66.48 % | 67.83 % |
| DEP_ReSCa | 70.76 % | 67.98% | 69.34 % |
| PAS_ReSCa,b | 74.83 % | 62.50 % | 68.11 % |
Totally, six different methods were implemented. The abbreviation DEP stands for the dependency-based graph kernel, PAS stands for the graph kernel based on predicate-argument-structure, SC stands for semantic class information, and ReSC stands for refined semantic class information. DEP_SC means that semantic class information is incorporated into the dependency-based graph kernel. Precision (P), Recall (R) and F-measure (F 1) were reported for each method. The highest performance under each evaluation criterion is bolded.
a means the performance difference between the underlying method and DEP is statistically significant
b means the performance difference between the underlying method and PAS is statistically significant. (p-value < 0.05)