MIT Press

Skip to Main Content

Table 4

Performance, in terms of precision (P), recall (R), and F1, of the CRF model on the development set for both event mention detection and event detection + classification tasks with different settings of features.

	MENTION DETECTION ONLY			DETECTION + CLASSIFICATION
	P	R	F1	P	R	F1
ALL FEATURES	86.60%	80.56%	83.33%	30.05%	25.24%	28.45%
- without lemma	87.92%	79.48%	83.19%	33.43%	19.19%	22.19%
- without PoS	85.01%	74.32%	78.83%	38.83%	22.98%	27.26%
- without genre	86.74%	80.60%	83.41%	41.00%	25.46%	29.13%
BASELINE (only tokens)	82.72%	71.53%	76.15%	36.29%	20.15%	24.56%

	MENTION DETECTION ONLY			DETECTION + CLASSIFICATION
	P	R	F1	P	R	F1
ALL FEATURES	86.60%	80.56%	83.33%	30.05%	25.24%	28.45%
- without lemma	87.92%	79.48%	83.19%	33.43%	19.19%	22.19%
- without PoS	85.01%	74.32%	78.83%	38.83%	22.98%	27.26%
- without genre	86.74%	80.60%	83.41%	41.00%	25.46%	29.13%
BASELINE (only tokens)	82.72%	71.53%	76.15%	36.29%	20.15%	24.56%