Abstract:
Recent research has demonstrated the potential of mining geotagged Twitter data in order to identify distinct places as spatial clusters of thematically congruent tweets posted from these locations. But social media interaction and participation is not only textual: social media platforms are multimedia in nature, encompassing imagery as well as text. Accordingly, a research question emerges on whether geotagged imagery posted in social media can also be analyzed to reveal thematic clusters, furthering our abilities to harvest platial content from such crowd-contributed content. Such studies can be enabled by the recent advent of convolutional neural networks that can be trained to automatically and accurately classify imagery. In this thesis we pursue a study of automatically classified crowd-contributed geotagged imagery from six major cities, in order to assess the emergence of spatial semantic associations.
