Table 3.

Geotagging F-Scores for GeoWebNews featuring the best performing models

NER model/geoparser	Precision	Recall	F-Score
NCRF++ (literal and associative labels)	79.9	75.4	77.6
Yahoo! placemaker	73.4	55.5	63.2
Edinburgh geoparser	81	52.4	63.6
SpacyNLP	82.4	68.6	74.9
Google cloud natural language	91.0	76.6	83.2
NCRF++ (“Location” label only)	90.0	87.2	88.6

The NCRF++ models’ scores were averaged over fivefolds ($\sigma$ = 1.2–1.3)