Measurement and Analysis of Online Social Networks By

Measurement and Analysis of Online Social Networks By Alan Mislove, Massimiliano Marcon, Krishna P. Gummadi, Peter Druschel, Bobby Bhattacharjee Attacked by Ionut Trestian

Goals of the paper (1) • Understanding social graphs for 2 things: – Improving current systems – Designing new applications Never happened • Confirm properties of online social networks (e. g. power law, small world, scale-free) Even the authors acknowledge that it has been shown in previous studies

Goals of the paper (2) • Detecting trusted or influential users ? • Mitigate email spam • Improve Internet search • Defend against Sybil attacks These are awesome goals, I agree. Why don’t you spend your time actually tackling them?

Goals of the paper (3) • Large scale? – If showing the same thing for a larger number of people is your main contribution then you could have written your paper in just a few lines. • More social networks?

More social networks ? • How about networks with stronger identity enforcements? • The networks that you have a strong user population from are mostly content based (e. g. You. Tube, Live. Journal, Flickr) (only 11% from Orkut)

Paper results (1) • Symmetric links – 62. 0% Flickr – 73. 5% Live. Journal – 100. 0% Orkut – 79. 1% You. Tube High degree of link symmetry Basically means that if you are friend with someone he’s a friend with you …

Paper results (2) Distributions of node indegree and outdegree are very similar Isn’t this a clear consequence of the high link symmetry ?

Paper results (3) Actually most results seem to be a consequence of the high link symmetry property

Six degrees of separation (1) • Classical result by Stanley Milgram • Showed that any two individuals are separated by an average of six acquaintances • Another not so well known result is that most users are connected through a very small core of influential users – the present paper calls them critical

Six degrees of separation (2) • In the real world these hubs are real people • In a social network they just represent bits stored on a hard drive 10110110110000010010101010000 • If you can mess with the bits that define a hub-type user you can mess with any of them • How are these critical?

Some final points • This study in no way seems to capture the actual dynamics of Social Networks • You actually note that you observed a big difference between datasets collected at close times (two months) • Even your future work on Ostra on leveraging thrust uses a not so suitable dataset that you acknowledge.

Conclusions • Your paper talks about random graphs • The whole paper seems random ! • Findings seem obvious and you acknowledge that they have been previously reported • It seems more useful that you would spend your time tackling the goals - mitigating email spam, improving Internet search defending against Sybil attacks