Table 9. The loss and confusion matrix results of the four USE flavors on the six datasets.

Keyword	Dataset	Loss	TP	TN	FP	FN
USE-M (Tuned)	Before Normalization	0.718	109	136	25	35
USE-MQA (Tuned)	Before Normalization	0.442	116	136	25	28
USE-M	Before Normalization	0.444	123	127	34	21
USE-MQA	Before Normalization	0.431	111	144	17	33
USE-M (Tuned)	After Normalization	1.063	111	118	43	33
USE-MQA (Tuned)	After Normalization	1.338	114	118	43	30
USE-M	After Normalization	0.538	117	112	49	27
USE-MQA	After Normalization	0.473	115	123	38	29
USE-M (Tuned)	Stemmed	0.750	115	132	29	29
USE-MQA (Tuned)	Stemmed	1.590	103	126	35	41
USE-M	Stemmed	0.527	113	111	50	31
USE-MQA	Stemmed	0.526	94	130	31	50
USE-M (Tuned)	Lemmatized	0.483	90	142	19	54
USE-MQA (Tuned)	Lemmatized	1.344	102	123	38	42
USE-M	Lemmatized	0.523	113	114	47	31
USE-MQA	Lemmatized	0.477	112	123	38	32
USE-M (Tuned)	Stemmed and Filtered	1.039	106	125	36	38
USE-MQA (Tuned)	Stemmed and Filtered	1.189	93	120	41	51
USE-M	Stemmed and Filtered	0.570	100	115	46	44
USE-MQA	Stemmed and Filtered	0.598	68	139	22	76
USE-M (Tuned)	Lemmatized and Filtered	1.102	101	123	38	43
USE-MQA (Tuned)	Lemmatized and Filtered	0.877	103	123	38	41
USE-M	Lemmatized and Filtered	0.618	104	121	40	40
USE-MQA	Lemmatized and Filtered	0.539	114	113	48	30