Transcript 7-Marcel-offense - Northwestern Networks Group

Offense: Planetary-Scale Views
on a Large Instant Messaging
Network
J. Leskovec, et al.
As a whole…
 Provides a strong analysis
 Considers a variety of approaches
 However, it falls prey to a number of
critical pitfalls
The Data Set
 Only includes data from MSN Messenger
 Other networks may have different usage
patterns
 Paper suggests MSN accounts for ~50% at the time
(seems to be correct)
 Each network could draw an entirely different
type of user (CompuServe vs. AOL)
 What about spam/bots?
 Are these a relevant portion?
Communication Demographics
 Only look at ages 10-60, to correct for any
misrepresented ages
 Any theoretical backing?
 Conversations depend only on number of
messages
 What about non-symmetric conversations? (one
user sending messages, another not replying)
World Geography and
Communication
 Show us distribution of MSN users by
World population
 What does this really tell us?
 Better to instead look per capita to computer
users?
 Perhaps internet use?
 Would take into account areas where no one is
on the internet and therefore not on MSN
The Communication Network
 Only consider those who communicated
during time period
 A limitation of the data set
 Doesn’t acknowledge the possibility of large
amounts of potential links that were just not
active during the month
 Really need a larger data set to deal with this
Milgram
 Data is not actually comparable to
Milgram’s!
 Milgram focused on drawing connections
with a letter
 These connections are more incidental (based
around some sort of social network, but are not
“focused” in the same way)
 Many people may be “on a first name basis”
with someone, but just not normally talk to
them
Milgram
 Temporally incomparability
 This data relies only on 1 month of
communication
 In Milgram’s study, people could use anyone
that they knew on a first name basis
 A whole lifetime!
 Implication: MSN may be much better
connected than these results show!
A Purpose?
 All this information is good to know
 But what can we really do with it?
 Generally speaking, seems people use it as
an extension of their real world networks
 Experiment like this unable to describe any
motivation, which would be helpful