Research / 2021 / Article / Tab 2

Research Article

Hierarchical Network with Label Embedding for Contextual Emotion Recognition

Table 2

Ablation experimental results with different training objectives.

MetricsM+C+LM+CM+LM

Micro-F1 (↑)0.56650.56550.55700.5539
Macro-F1(↑)0.41860.42460.41560.4128
Ranking loss (↓)0.11320.12090.11940.1272
One-error (↓)0.35590.37340.37190.3787
Coverage (↓)2.12722.17782.16382.2234
Hamming loss (↓)0.19980.19590.20400.1957

“M”: multilabel loss; “C”: cosine embedding loss; “L”: label embedding loss.