Analyzing image-based political propaganda in referendum campaigns: from elements to strategies

EPJ Data Science

Table 2 Vision API text verification result.

Text	TP	FP	FN	Precision	Recall
(a) Vision API OCR test in pan-blue images
Referendum	355	64	187	0.8472	0.6549
KMT	174	21	86	0.8923	0.6692
Taiwan	144	42	128	0.7741	0.5294
1218¹	132	57	41	0.6984	0.7630
Vote	111	21	32	0.8409	0.7762
Disagree	74	17	51	0.8131	0.5920
Racto. pork²	73	3	199	0.9605	0.2683
Algal reef	65	8	42	0.8904	0.6074
Government	63	16	10	0.7974	0.8630
Health	43	20	9	0.6825	0.8269
(b) Vision API OCR test in DPP (pan-green) images
Referendum	1036	47	232	0.9566	0.8170
1218¹	412	64	133	0.8655	0.7559
4 disagree	420	16	137	0.9633	0.7540
Disagree	426	100	56	0.8098	0.8838
TW powerful⁵	348	13	77	0.9639	0.8188
Fourth NPP³	269	20	107	0.9307	0.7154
Taiwan	231	145	16	0.6143	0.9352
NGT⁴	220	10	77	0.9565	0.7404
Pork from US	184	10	48	0.9484	0.7931
Vote	182	46	5	0.7982	0.9732