Table 3 10-fold cross validation performance on different feature sets combinations. (Feature sets: (a) Word n-grams; (b) POS tags; (c) Clusters; F: F-1 score; P: precision; R: recall; for the categories that do not indicate the metric, F-1 score are used)
Feature sets | (a) | (a) + (b) | (a) + (c) | (a) + (b) + (c) | |
|---|---|---|---|---|---|
Micro-averaging | F | 0.7208 | 0.7263 | 0.7255 | 0.73 |
Macro-averaging | P | 0.5402 | 0.5438 | 0.5396 | 0.5477 |
R | 0.4386 | 0.4468 | 0.4442 | 0.4576 | |
F | 0.4841 | 0.4905 | 0.4872 | 0.4986 | |
Unrelated | 0.8599 | 0.864 | 0.859 | 0.8618 | |
Neutral | 0.6181 | 0.6226 | 0.625 | 0.6231 | |
Positive | 0.7021 | 0.7098 | 0.7123 | 0.7136 | |
NegSafety | 0.7277 | 0.734 | 0.7357 | 0.7542 | |
NegEfficacy | 0.2593 | 0.3214 | 0.2593 | 0.3793 | |
NegCost | 0 | 0 | 0 | 0 | |
NegResistant | 0 | 0 | 0 | 0 | |
NegOthers | 0.4645 | 0.4614 | 0.4724 | 0.4753 | |