Table 2.

Performance metrics of the Amazon Rekognition service. The first column presents a tag identifier and number (N) of unique images that were tagged. Following this are performance metrics used in pattern recognition: true positives (TP), true negative (TN), false positives (FP), false negatives (FN), precision (proportion of correctly tagged images), recall (proportion of cases associated to the tag that were correctly retrieved), and F-score (the harmonic mean of precision and recall). We also indicate the metric values obtained after correcting the FP in brackets. The face-ism row evaluates the cases in which the face and person was correctly detected and associated (i.e. the face tag being inside the body tag).

Tag (N)	TP	TN	FP	FN	Precision	Recall	F-score
Face (4039)	3229	794 (802)	8 (0)	56	.998 (1)	.983	.990 (.991)
Face-ism (3237)	3137	44 (49)	5 (0)	51	.998 (1)	.984	.991 (.992)
Female face (3237)	1483	1577 (1702)	125 (0)	52	.922 (1)	.966	.944 (.983)
Male face (3237)	1630	1427 (1535)	108 (0)	72	.938 (1)	.958	.948 (.978)