Predicting Mental Health Problems with Automatic Identification of Metaphors

Table 4

Prediction performance on the eRisk 2017 dataset.

Method		Accuracy	F1-score
Trotzek et al.	CNN	0.88	0.59
	LR	0.88	0.69
MSM	Sentiment	0.81	0.61
	Metaphor	0.87	0.56
	ALL	0.89	0.70

All: sentiment + metaphor.