|The first step towards efficient social media content analysis is to understand it and identify means of user interaction. Trying to study the problem from the user perspective, we analyze user-generated photos uploaded to famous Flickr social network, in order to extract meaningful semantic trends covering specific research aspects, like content popularity, spatial areas of interest and popular events. Initially, we select a geographical area of social interest, like a city center, defined by a strict bounding box. We then cluster photos taken within the box based on their geo-tagging metadata information (i.e., their latitude and longitude information) and divide large areas into smaller groups of fixed size, which we will refer to as “geo-clusters”. Within these geo-clusters, we further identify semantically meaningful “places” of user interest, by analyzing any additional textual metadata available, i.e., user selected tags that characterize each place's photos. By post-processing the latter, we are then able to rank them and thus select the most appropriate tags that describe landmarks and other places of interest, as well as events occurring within these places of interest. As a next step, we place these tags on a map and help users to intuitively visualize places of interest and the actual photo content at a glance. Finally, we examine the temporal dynamics of analyzed photos over a long period of time, so as to obtain the underlying trends to be identified within this kind of social media generated content.
|E. Spyrou, Ph. Mylonas , "Analyzing Flickr metadata to extract location-based information and semantically organize its photo content", Neurocomputing, Elsevier, Volume 172, pp. 114–133, January 2016|