1: \begin{abstract}

2: We propose efficient algorithms for two key tasks in

3: the analysis of large nonuniform networks: uniform node

4: sampling and cluster detection.

5: Our sampling technique is based on augmenting a simple, but

6: slowly mixing uniform MCMC sampler with a regular random

7: walk in order to speed up its convergence; however the

8: combined MCMC chain is then only sampled when it is in its 

9: ``uniform sampling'' mode.

10: Our clustering algorithm determines the relevant neighbourhood

11: of a given node $u$ in the network by first estimating the Fiedler

12: vector of a Dirichlet matrix with $u$ fixed at zero potential,

13: and then finding the neighbourhood of $u$ that yields a minimal

14: weighted Cheeger ratio, where the edge weights are determined by

15: differences in the estimated node potentials.

16: Both of our algorithms are based on local computations, i.e.\

17: operations on the full adjacency matrix of the network are not used. 

18: The algorithms are evaluated experimentally using three types of

19: nonuniform networks: Dorogovtsev-Goltsev-Mendes

20: ``pseudofractal~graphs'', scientific collaboration networks, 

21: and randomised ``caveman graphs''.

22: 

23: \end{abstract}

24: