Security Informatics

Table 9 Results for improvements on our PAN2012 approach

	*Filter*	*Categories*	CT	TP	FP	FN	*Precision*	*Recall*	F1
Results from Train Corpus based on highest recall
1.	Original	All 4	1	113	294	29	0.28	0.8	0.41
2.	Improved -no Filter	All 4	1	134	683	8	0.16	0.94	0.28
3.	Computer	All 4	1	131	529	11	0.2	0.92	0.33
4.	Sex	All 4	1	130	540	12	0.19	0.92	0.32
5.	Combined	All 4	1	127	399	15	0.24	0.89	0.38
Results from Train Corpus based on highest F1
6.	Original	All 4	2	88	38	54	0.7	0.62	0.66
7.	Improved - no Filter	All 4	3	114	53	28	0.68	0.8	0.74
8.	Computer	All 4	3	111	34	31	0.77	0.78	0.77
9.	Sex	All 4	3	99	47	43	0.68	0.7	0.69
10.	Combo	All 4	3	98	30	44	0.77	0.69	0.73
Results from Test Corpus based on highest F1
11.	Original	All 4	2	99	61	155	0.62	0.39	0.48
12.	Computer	All 4	3	115	57	139	0.67	0.45	0.54