Navigation in Networks, Revisited Networked Life MKSE 112 Fall

Navigation in Networks, Revisited Networked Life MKSE 112 Fall 2012 Prof. Michael Kearns

Lecture Roadmap Distinguish between structural and algorithmic aspects of navigation Introduce another local/long-distance formation model Investigate the two aspects in this model Compare to some real-world data

Two Aspects of Navigation In order for people (or machines) to find short paths in networks: – short paths must exist (structural; small diameter) – people must be able to find the short paths via only local forwarding (algorithmic) The algorithmic constraints are strong (Travers & Milgram) – only know your neighbors in the network – limited information about the target/destination (physical location, some background) Look at a model incorporating structural and algorithmic constraints

Kleinberg’s Model Start with an k by k grid of vertices (so N k 2) – each vertex connected to compass neighbors – add a few random ”long-distance” connections to each vertex – probability p(d) of connecting to a vertex at grid distance d: – large r: heavy bias towards “more local” long-distance connections – small r: approach uniformly random p(d) (1/d) r ,r 0

Which values of r: permit efficient navigation? Efficient: number of hops N, e.g. log(N) Algorithmic assumption: – – – – – Kleinberg’s Question p(d) (1/d) r ,r 0 vertices know the grid addresses of their neighbors vertices know the grid address of the target (Sharon, MA) vertices always forward the message to neighbor closest to the target in grid distance no “backwards” steps, even if helpful purely geographic information

Kleinberg’s Result Intuition: – if r is too large (strong local bias), then “long-distance” connections never help much; short paths may not even exist – if r is too small (no local bias), we may quickly get close to the target; but then we’ll have to use grid links to finish – effective search requires a delicate mixture of link distances The result (informally): as N becomes large: – r 2 is the only value that permits rapid navigation ( log(N) steps) – a “knife’s edge” result; very sensitive Note: locality of information crucial to this argument – At r 2, will have small diameter, but local algorithms can’t find the short paths

From Brockmann, Hufnagel, Geisel (2006) Best-fit value of r 1.59

Navigation via Identity Watts et al.: – we don’t navigate social networks by purely “geographic” information – we don’t use any single criterion; recall Dodds et al. on Columbia SW – different criteria used at different points in the chain Represent individuals by a vector of attributes – – – – profession, religion, hobbies, education, background, etc attribute values have distances between them (tree-structured) distance between individuals: minimum distance in any attribute only need one thing in common to be close! Algorithm: – given attribute vector of target – forward message to neighbor closest to target Let’s look a bit at the paper Permits fast navigation under broad conditions – not as sensitive as Kleinberg’s model all jobs scientists CS athletes chemistry tennis baseball

Summary Efficient navigation has both structural and algorithmic requirements Kleinberg’s model and question captures both Result predicts delicate mixture of connectivity for success Not too far from reality? (Where’s George? data) Watts et al. provide more “sociological” model More complex, but less sensitive