Social Media Data for Content Creation in Location-based Games

Geo-tagged images from social media services (Flickr, Foursquare, Yelp and Google Places) are extracted from 2019 - 2022 to serve content creation purpose in location-based games. Images are evaluated for their representativeness, postional accuracy and relevant tag. The experiments with six worldwide regions show that the Flickr images are representative of place with higher probability (32%) and location accuracy (96%), but relevant name extraction relied on the external method (Tag-tag). Foursquare and Yelp images always have the correct location (100%), and the name is usually relevant (100%), but the image is rarely representative (22%, 4%). Google Places, despite providing the large volumes of data and a variety of establishments, only 5% of representative images were found; however, location accuracy (75%) and relevant names (100%) are reasonable measures.

[Download Dataset]