Preview

PDF, English Download (23MB) | Terms of use

Abstract

Given our evermore complex world, keeping track of important events, developments, and interdependencies is increasingly time-consuming or even infeasible. To a large part, this complexity is caused by the systems' underlying dynamics and connectedness – characteristics found in various domains. Media, characterized by its high volatility and interwoven network of content and actors, is such a domain. Due to its ever-growing importance, not only from an academic but also from a societal perspective, understanding media-related phenomena is of huge importance. However, deriving insights from media data is not trivial, and the mentioned characteristics must be considered. Instead of simply asking a question like "Which topics are currently discussed?", to gain a holistic perspective, one must also consider the underlying dynamics and ask "Which topics are gaining in popularity?" or "How does the relevance of a topic change over time?". Similarly, it is insufficient to only investigate individual media actors without considering their social connectedness. To account for these requirements, in this thesis, we leverage temporal, network-based methods for analyzing media data. We do not limit ourselves to the development of novel methodological approaches but also put these into practice by bridging the gap between model and application.

First, a unifying model that allows for coping with heterogeneous data sources is required. Based on the concept of temporal networks, we develop such a model that particularly reflects the data's time-sensitivity and structural interdependencies. Its capabilities are demonstrated in several media analytics studies. These studies include the investigation of trends – a phenomenon that is integral to the dynamics of media. In particular, we focus on examining long-term trends, which keep their large prevalence over longer periods compared to short-lived trends. Also, we connect the analysis of trends with that of social network analysis by investigating the actor-networks underlying trends. Further, we show that the model can also be applied to online conversation analysis. Given its conception based on temporal networks, we can approach respective analysis by incorporating the conversations' content, dynamics, and structural properties. Finally, after laying the theoretical foundation of this work in the form of the proposed model and successfully leveraging it for several analytics use cases, we shift our focus to its technical implementation. For that, we showcase two real-world applications, the EPINetz platform and the TrendTracker app, for the temporal and network-based exploration of media data. Also, the design of such applications at its core requires a performant graph data management and analysis system. Therefore, we benchmark various system setups and discuss an appropriate implementation strategy. In sum, this thesis demonstrates the benefits that come with approaching media analysis from a temporal and network-based perspective. Our contributions are not limited to novel methods and techniques but also tackle challenges that occur when putting these approaches into practice.