Computational Social Network Analysis of Authority in the Blogosphere

Social Media have gained more and more importance in many areas of our daily lives. One of the first media types in this field were weblogs, which allow everyone to easily publish content online. For weblogs, the reliable algorithmic detection of importance based on social reputation is still an open issue. In this thesis we attempt to measure this authority with algorithms from the field of Social Network Analysis, which have to be scalable, transparent and thoroughly evaluated. Social scientists have identified very specific characteristics for the elite group of influential tob bloggers, which are well represented by the network core/periphery model from Borgatti & Everett. We approximate this model with a scalable algorithm based on the concept of k-cores from Seidman. For evaluation we collect datasets of thousands of top blogs in six different languages, in order to compare and crosscheck the results. These are also compared to random networks, in order to show the significance of the findings. Remaining detection problems are engaged with anomaly detection and network filtering algorithms, which lead to an overall reliable detection process according to our evaluations. In a second step, this thesis transfers these insights to a practical problem. A complete mining and analysis methodology for the monitoring of specific entities in the blogosphere is developed and evaluated. It consists of the search for relevant blog articles, which proves to be highly effective, and the authority measurement of these articles for potential end users in business scenarios, which are validated with respect to soundness. The resulting tool, the “Social Media Miner”, integrates this methodology, combined with text processing methods, in an extensive analysis process and received very good feedback.