Table 4. Experiments results.

Data set	Classifier	BoW		LDA		Improv.
		Mean	Sdev.	Mean	Sdev.
Ohsumed C04	SVM	0.73	2.359 ⋅ 10−2	0.67	2.569 ⋅ 10−2	NO
	k-NN	0.15	4.945 ⋅ 10−2	0.61	2.867 ⋅ 10−2	YES
	Bayes	0.61	2.547 ⋅ 10−2	0.28	1.89 ⋅ 10−2	NO
Reuters	SVM	0.90	1.233 ⋅ 10−2	0.91	1.214 ⋅ 10−2	–
	k-NN	0.51	3.143 ⋅ 10−2	0.89	1.370 ⋅ 10−2	YES
	Bayes	0.68	1.928 ⋅ 10−2	0.63	1.639 ⋅ 10−2	NO
20News	SVM	0.75	1.057 ⋅ 10−2	0.72	9.15 ⋅ 10−3	NO
	k-NN	0.42	1.139 ⋅ 10−2	0.68	9.850 ⋅ 10−3	YES
	Bayes	0.51	1.310 ⋅ 10−2	0.49	1.503 ⋅ 10−2	NO
TREC Allele	SVM	0.58	5.336 ⋅ 10−2	0.47	5.631 ⋅ 10−2	NO
	k-NN	0.00	0.000	0.430	5.755 ⋅ 10−2	YES
	Bayes	0.46	3.000 ⋅ 10−2	0.09	1.226 ⋅ 10−2	NO
Yelp Polarity	SVM	0.67	2.687 ⋅ 10−2	0.49	2.745 ⋅ 10−2	NO
	k-NN	0.24	3.574 ⋅ 10−2	0.42	3.350 ⋅ 10−2	YES
	Bayes	0.57	3.690 ⋅ 10−2	0.41	3.334 ⋅ 10−2	NO
Yahoo!	SVM	0.43	1.549 ⋅ 10−2	0.47	1.987 ⋅ 10−2	YES
	k-NN	0.04	1.535 ⋅ 10−2	0.38	2.222 ⋅ 10−2	YES
	Bayes	0.43	1.816 ⋅ 10−2	0.31	1.803 ⋅ 10−2	NO

Results obtained.