Characterization and Analysis of Emergent Image Semantics Using Network Models

Understanding and dealing with the emergent semantics of image and media-based information is one of the most challenging aspects of theoretical, algorithmic, and systems-oriented research in Multimedia. Emergent semantics implies that media is endowed with meaning by placing it in context of other similar media and through factors that are user specific. This means that unlike alphanumeric data, a fixed semantics cannot be assigned to media. While this intriguing property of media-based information has been known for nearly a decade, progress towards development of rigorous frameworks to represent and analyze this phenomenon has been limited. In this paper, we present results that move towards addressing this problem. Specifically, we show how the emergent semantics of a data collection can be first formalized and then captured and represented using network models across users. Using real-world data from a group of users, we then show how such networks can be theoretically characterized and highlight many of their important properties. The primary results communicated in this paper include: (1) a graph-theoretic approach for formalization of the notion of emergent semantics, (2) description of how real-world emergent semantics can be captured and represented as networks, and (3) investigation of the issue of quantitative characterization of emergent semantics through the analysis of these networks.