Table 4.

Performance comparisons on NCBI and CDR corpora

Corpus	Method	P	R	F
NCBI	BANNER [4]	83.80	80.00	81.80
	Bi-LSTM + WE [14]	84.87	74.11	79.13
	MCNN	85.08	85.26	85.17
	MCNN*	83.74	83.03	83.39
CDR	HITSZ_CDR [5]	88.68	85.23	86.93
	Lee et al.’s [6]	87.34	83.75	85.51
	CRD-DNER [31]	79.49	73.58	76.42
	MCNN	88.20	87.46	87.83

Note. MCNN* is the version of removing the lexicon feature embedding and the post-processing step