ABSTRACT
Digital media, like images, videos, etc., play an important role in social networks. For example, users in social networks are able to post images depicting themselves and/or some of their friends in a location, while other users may see some of these images and rate them. Such rates can be used to describe connections between users or between users and locations, etc. Thus, digital media content analysis is of particular interest in social networks. Digital media representations are usually high-dimensional. For example, a facial image of size 200× 150 pixels can be represented by a 30000-dimensional vector obtained by using each pixel coordinate as a different dimension. Therefore, significant efforts have been devoted to deriving low-dimensional data representations that retain properties of interest of the data,
like pair-wise distances, data dispersion, and class discrimination [YXZ+07, RR13]. Such low-dimensional data representations are essential in order to reduce the computational cost and the physical memory used to store data, especially in social network applications, where the cardinality of the data sets is enormous. In addition, machine learning methods that are able to classify digital media data are required, in order to proceed with automatic data categorization toward decision making and/or recommending appropriate digital media content to users [KZP06].
