Table 4.

Performance of models with FCD features on named entity classification and recognition

ID	Feature(model)	Classification (all terms)			Classification (OOV terms)			Named entity recognition
		Precision	Recall	F-score	Precision	Recall	F-score	Precision	Recall	F-score
Run 1	Lexical (linear)	75.52	85.63	80.26	74.03	75.68	74.85	85.70	78.36	81.86
Run 2	FCD (linear)	81.59	87.77	84.57 (+4.31)	83.74	87.64	85.64 (+10.79)	87.98	80.70	84.18 (+2.32)
Run 3	FCD (SVD + RBF)	83.02	88.24	85.55 (+5.29)	83.12	85.31	84.2 (+9.35)	89.80	81.76	85.59 (+3.73)
Run 4	FCD (Combine (2, 3))	82.46	90.35	86.23 (+5.97)	83.21	88.35	85.7 (+10.85)	89.29	82.45	85.74 (+3.88)
Run 5	All (linear)	82.96	89.31	86.02 (+5.76)	83.65	89.16	86.32 (+11.47)	89.93	81.71	85.62 (+3.76)
Run 6	All (Combine (3, 5))	83.94	89.99	86.86 (+6.6)	83.92	88.86	86.32 (+11.47)	90.37	82.40	86.20 (+4.34)

In Run 1, 2 and 5 SVMs with linear kernel are used. In Run 3, SVD is used to reduce the feature dimension and a SVM with RBF kernel is used to classify examples. In Run 3 only features related to CDF I are used. In Run 4 outputs of Run 2 and 3 are combined. Run 6 is the combination of Run 3 and Run 5.