Second Space: A Generative Model for the Blogosphere

Analysing complex natural phenomena often requires syn- thesized data that matches observed characteristics. Graph models are widely used in analyzing the Web in general, but are less suitable for modeling the Blogosphere. While blog networks resemble many properties of Web graphs, the dy- namic nature of the Blogosphere, its unique structure and the evolution of the link structure due to blog readership and social interactions are not captured by the existing models. We describe an agent-based simulation model to construct blog graphs that exhibit properties similar to the real world blog networks in their degree distributions, degree correla- tion, clustering coefficient and reciprocity. The model can help researchers analyze the Blogosphere and facilitates the development and testing of new algorithms. We capture the linking patterns arising in the Blogosphere through local interactions, as suggested in (Vazquez 2003). These local interactions comprise the interactions of a blog with other blogs connected to it either by inlinks or outlinks. Key properties of the resulting graph include the degree dis- tributions, degree correlation, clustering coefficient, average degree, reciprocity and the distribution of connected compo- nents. To the best of our knowledge, there exist no general models to generate the blog and post networks which pos- sess the properties observed in the real world blogs. Table 1 gives a quick comparison of the existing Web model and shows the need for a model for Blogosphere. Design. The observed characteristics of bloggers effect the structure of the two graphs underlying the Blogosphere as shown in Figure 1. The blog networkis a network of blogs obtained by collapsing all directed post links between blog posts into directed edges between blogs. Blog networks give a macroscopic view of the Blogosphere and help to infer a social network structure, under the assumption that blogs that are part of the same community link each other more often. The post network (Leskovec et al. 2007) is formed by ignoring the posts' parent blogs and focusing on the link structure among posts only. Post networks give a micro- scopic view of the Blogosphere with details like which post linked to which other post and at what time.