Table 4.

Topic classification performance measured by F score on different topic categories, with five classifiers. bsline: the system simply tags all sentences with all 11 labels; l-lda: the labeled LDA classifier; svm: the SVM classifier using bag of words as features; svm-e: the SVM classifier using word embedding as features; cnn: the convolutional neural network classifier

	bsline	l-lda	svm	svm-e	cnn
Micro	19.3	54.4	55.8	58.3	65.4
ALTR	6.5	9.2	9.4	30.7	35.5
DAIL	12.5	30.1	28.8	46.4	48.1
DIAG	22.2	58.8	60.2	65.3	67.1
FIND	23.4	50.1	50.9	60.0	60.3
HSYS	17.5	45.4	41.1	55.3	57.7
MISC	35.7	76.2	75.8	71.4	78.1
NUTR	12.6	57.3	58.6	68.4	72.8
PERS	20.2	24.4	26.5	47.7	47.8
RSRC	11.9	48.0	48.3	55.2	61.1
TEST	6.3	27.6	26.1	47.9	52.6
TREA	37.5	65.7	66.0	64.2	73.6