Conceptualization of place via spatial clustering and co-occurrence analysis

More and more users are contributing and sharing more and more contents on the Web via the use of content hosting sites and social media services. These user-generated contents are tagged with terms characterizing the contents from the users' perspectives. Massive collections of tagged photos in popular photo hosting sites are well known for their richness in semantic extent and geospatial scope. Furthermore, geo-tags, which are machine-generated positional data, are frequently embedded within these photos. We develop in this paper an approach based on the analyses of tags and geo-tags for the exploration and characterization of the implicit localities in collections of user photos. At the same time, the approach also allows us to explore the meanings given by users about the places in their photo collections. In this approach, we first use DBSCAN (Density-based Spatial Clustering with Noise) to group geo-tagged photos into clusters (of possibly multiple distance scales). Then, a co-occurrence analysis on the tags used within a cluster is utilized to extract conceptualization of the place in the cluster. The extracted concepts are not necessarily of geospatial nature (e.g., airplane/airline names in photos taken in the surrounding area of an airport) so are especially useful when compared to concepts extracted via the simple use of readily available locational resources (e.g., gazetteers).