A Kernel-Based Approach for Biomedical Named Entity Recognition

Table 4

Our system compared with existing systems.

System	ML approach	Domain knowledge	F-score%
Normal SVM	Linear SVM	—	56.09%
Our system	SVM with SL	—	60.60%
Zhou and Su (2004) [2] final	HMM, SVM	Resolution of name alias, cascaded NEs, and Abbreviations; dictionary; POS	72.55%
Zhou and Su (2004) [2]	HMM, SVM	(baseline)	64.1%
Song et al. (2004) [15] final	SVM, CRF	POS information, phrase, and virtual sample	66.28%
Song et al. (2004) [15]	SVM	(baseline)	63.85%
Saha et al. (2010) [18] final	Composite kernel	—	67.89%