Modelling Confidence in Extraction of Place Tags from Flickr

The volume and potential value of user generated c ontent is ever growing. One such valuable source for better understanding of naïve o r vernacular geography is in the form of geotagged images from Flickr. Research in the past has looked into automatic identification of place tags from this source. This paper gives an overview of a data mining techniques for identification of places tag s at different levels of detail and a Bayesian infere nce model to predict the probability for each selec t d tag as being ‘non-noise’.