Most projects involving structural mining of Internet data have
focused on properties of either the topology of the physical network
(i.e., the actual routers and smaller networks that make up the
Internet) or the link structure of the World Wide Web. These studies
have greatly enhanced our understanding of the structure and function
of the Internet and the Web, but they do not capture much information
about how these networks are actually used; to do this, we must extend
our analysis to large volumes of usage data obtained from the network.
This talk will focus on the challenges involved in gathering
usage-based network data sets while preserving user privacy, reducing
them to a manageable size, and mining them for useful insights into
user behavior. Two data sets will be discussed at length: (1) network
flow data from the Internet2 network, and its use in modeling user
behavior and constructing a taxonomy of network applications; and (2)
Web request data from the Indiana University network, and its use in
improving the random-surfer model of Web traffic used by link-based
algorithms such as PageRank.
Audio (MP3 File, Podcast Ready)
Copyright. All Rights Reserved.