Second Space: A Generative Model For The Blogosphere

1
Second Space A Generative Model For The
Blogosphere
Amit Karandikar, Akshay Java, Anupam Joshi, Tim
Finin, Yelena Yesha, and Yaacov Yesha
Goal generate graph structures better
approximating those found on the Blogosphere
Motivation To test the new algorithms for
Blogosphere, save time and effort in gathering
and preprocessing the "real" data, extrapolate
the properties by varying parameters
Graphs are everywhere ..
.. and so are Power Laws!
20 of the population owns 80 of the wealth.
Real networks often tend to show the rich get
richer phenomenon. Already popular website is
bound to get more inlinks.

• Idea simulate characteristics of Blogging agents
• Blog writers are enthusiastic blog readers
• Most bloggers post infrequently
• Active bloggers follow popular blogs, friends
  blogs and interact online.

Scale-free networks

• Ok, power laws
• So whats the big deal?
• Makes ranking of web content possible
• (e.g. Google ranking)
• Shows that only a few things are more important
  than others

The likelihood of linking to a popular website is
higher
Preferential Attachment

• Approach
• Model bloggers with read, write and idle states
• Select blog writers preferentially based on the
  outdegree of the blog node
• Perform preferential random walk in the blog
  neighborhood based on the neighbors indegree
• With a small probability perform totally random
  reading and writer selection

Conclusion
We better approximated Blogosphere graph
properties including degree distributions,
average shortest path, diameter, degree
correlations, reciprocity, size of connected
components