MIT Press

Table 6

Average precision (P), recall (R), and F1 over three runs of the BiLSTM system with the configuration suggested by Reimers and Gurevych (2017a).

TASK	P	R	F1
MENTION DETECTION ONLY	82.50%	83.53%	82.99%
DETECTION + CLASSIFICATION	63.46%	62.93%	63.19%