Table 5.

Comparing F₁-score of traditional models (TREC-6 dataset) with focusing POS tags (N+V+Adj+Det: NVAD).

		Rang of N-gram
		Micro $F_1$-score					Macro $F_1$-score
Dataset/Model		Unigram	Uni+ Bigram	Uni+ Trigram	Bigram	Trigram	Unigram	Uni+ Bigram	Uni+ Trigram	Bigram	Trigram
Original	MNB	0.9600	0.9660	0.9660	0.9640	0.9200	0.9538	0.9598	0.9599	0.9676	0.9249
	SVM	0.9640	0.9680	0.9720	0.9580	0.9200	0.9672	0.9709	0.9745	0.9619	0.9239
	LR	0.9600	0.9700	0.9680	0.9460	0.8980	0.9637	0.9728	0.9711	0.9512	0.9044
+all POS Tag	MNB	0.9560	0.9480	0.9380	0.9280	0.8960	0.9159	0.9099	0.9154	0.9058	0.8781
	SVM	0.9800	0.9780	0.9760	0.9640	0.9360	0.9821	0.9801	0.9780	0.9674	0.9397
	LR	0.9680	0.9740	0.9720	0.9560	0.8920	0.9711	0.9769	0.9741	0.9602	0.9028
+focusing Tags	MNB	0.9540	0.9620	0.9540	0.9460	0.9040	0.9268	0.9468	0.9391	0.9332	0.8954
	SVM	0.9660	0.9780	0.9700	0.9680	0.9320	0.9687	0.9799	0.9730	0.9708	0.9362
	LR	0.9620	0.9680	0.9620	0.9440	0.8940	0.9656	0.9710	0.9653	0.9496	0.9019
+focusing Words	MNB	0.9560	0.9660	0.9680	0.9500	0.9200	0.9503	0.9600	0.9615	0.9534	0.9251
	SVM	0.9720	0.9720	0.9680	0.9660	0.9300	0.9747	0.9747	0.9712	0.9693	0.9314
	LR	0.9480	0.9580	0.9580	0.9500	0.8880	0.9538	0.9621	0.9621	0.9549	0.8908