1. Data Mining for Complex Network Introduction and Background

2. Welcome!  Instructor: Ruoming Jin  Homepage: www.cs.kent.edu/~jin/www.cs.kent.edu/~jin/  Office: 264 MCS Building  Email: jin@cs.kent.edujin@cs.kent.edu

3. Course overview  The course goal  First of all, this is a research course, or a special topic course. There is no textbook and even no definition what topics are supposed to under the course name  Discussing the state-of-are techniques for mining complex networks  Review & Course Project

4. Simple Concepts (Review)  Erdős–Rényi model Random Graph Model  Markov Chain and Random Walk  Maximal Likelihood/Model Selection

5. Requirement  Each of you will have one presentation  Review Paper  Select a topic and review at least four papers  Bonus: Develop a new idea on this topic  Course Project  One or Two a group  Collect data; preprocess data; analyzing the data;  Final Grade: 30% presentations, 25% review, 35% project, and 10% class participation

6. What is a network?  Network: a collection of entities that are interconnected with links.  people that are friends  computers that are interconnected  web pages that point to each other  proteins that interact

7. Graphs  In mathematics, networks are called graphs, the entities are nodes, and the links are edges  Graph theory starts in the 18th century, with Leonhard Euler  The problem of Königsberg bridges  Since then graphs have been studied extensively.

8. Networks in the past  Graphs have been used in the past to model existing networks (e.g., networks of highways, social networks)  usually these networks were small  network can be studied visual inspection can reveal a lot of information

9. Networks now  More and larger networks appear  Products of technological advancement e.g., Internet, Web  Result of our ability to collect more, better, and more complex data e.g., gene regulatory networks  Networks of thousands, millions, or billions of nodes  impossible to visualize

10. The internet map

11. Understanding large graphs  What are the statistics of real life networks?  Can we explain how the networks were generated?  What else? A still very young field! (What is the basic principles and what those principle will mean?)

12. Measuring network properties  Around 1999  Watts and Strogatz, Dynamics and small- world phenomenon  Faloutsos 3, On power-law relationships of the Internet Topology  Kleinberg et al., The Web as a graph  Barabasi and Albert, The emergence of scaling in real networks

13. Real network properties  Most nodes have only a small number of neighbors (degree), but there are some nodes with very high degree (power-law degree distribution)  scale-free networks  If a node x is connected to y and z, then y and z are likely to be connected  high clustering coefficient  Most nodes are just a few edges away on average.  small world networks  Networks from very diverse areas (from internet to biological networks) have similar properties  Is it possible that there is a unifying underlying generative process?

14. Generating random graphs  Classic graph theory model (Erdös-Renyi)  each edge is generated independently with probability p  Very well studied model but:  most vertices have about the same degree  the probability of two nodes being linked is independent of whether they share a neighbor  the average paths are short

15. Modeling real networks  Real life networks are not “random”  Can we define a model that generates graphs with statistical properties similar to those in real life?  a flurry of models for random graphs

16. Processes on networks  Why is it important to understand the structure of networks?  Epidemiology: Viruses propagate much faster in scale-free networks  Vaccination of random nodes does not work, but targeted vaccination is very effective

17. The future of networks  Networks seem to be here to stay  More and more systems are modeled as networks  Scientists from various disciplines are working on networks (physicists, computer scientists, mathematicians, biologists, sociologist, economists)  There are many questions to understand.

18. Basic Mathematical Tools  Graph theory  Probability theory  Linear Algebra

19. Graph Theory  Graph G=(V,E)  V = set of vertices  E = set of edges 1 2 3 4 5 undirected graph E={(1,2),(1,3),(2,3),(3,4),(4,5)}

20. Graph Theory  Graph G=(V,E)  V = set of vertices  E = set of edges 1 2 3 4 5 directed graph E={‹1,2›, ‹2,1› ‹1,3›, ‹3,2›, ‹3,4›, ‹4,5›}

21. Undirected graph 1 2 3 4 5  degree d(i) of node i  number of edges incident on node i  degree sequence  [d(1),d(2),d(3),d(4),d(5)]  [2,2,2,1,1]  degree distribution  [(1,2),(2,3)]

22. Directed Graph 1 2 3 4 5  in-degree d in (i) of node i  number of edges pointing to node i  out-degree d out (i) of node i  number of edges leaving node i  in-degree sequence  [1,2,1,1,1]  out-degree sequence  [2,1,2,1,0]

23. Paths  Path from node i to node j: a sequence of edges (directed or undirected from node i to node j)  path length: number of edges on the path  nodes i and j are connected  cycle: a path that starts and ends at the same node 1 2 3 4 5 1 2 3 4 5

24. Shortest Paths  Shortest Path from node i to node j  also known as BFS path, or geodesic path 1 2 3 4 5 1 2 3 4 5

25. Diameter  The longest shortest path in the graph 1 2 3 4 5 1 2 3 4 5

26. Undirected graph 1 2 3 4 5  Connected graph: a graph where there every pair of nodes is connected  Disconnected graph: a graph that is not connected  Connected Components: subsets of vertices that are connected

27. Fully Connected Graph  Clique K n  A graph that has all possible n(n-1)/2 edges 1 2 3 4 5

28. Directed Graph 1 2 3 4 5  Strongly connected graph: there exists a path from every i to every j  Weakly connected graph: If edges are made to be undirected the graph is connected

29. Subgraphs 1 2 3 4 5  Subgraph: Given V’  V, and E’  E, the graph G’=(V’,E’) is a subgraph of G.  Induced subgraph: Given V’  V, let E’  E is the set of all edges between the nodes in V’. The graph G’=(V’,E’), is an induced subgraph of G

30. Trees  Connected Undirected graphs without cycles 1 2 3 4 5

31. Bipartite graphs  Graphs where the set V can be partitioned into two sets L and R, such that all edges are between nodes in L and R, and there is no edge within L or R

32. Linear Algebra  Adjacency Matrix  symmetric matrix for undirected graphs 1 2 3 4 5

33. Linear Algebra  Adjacency Matrix  unsymmetric matrix for undirected graphs 1 2 3 4 5

34. Random Walks  Start from a node, and follow links uniformly at random.  Stationary distribution: The fraction of times that you visit node i, as the number of steps of the random walk approaches infinity  if the graph is strongly connected, the stationary distribution converges to a unique vector.

35. Random Walks  stationary distribution: principal left eigenvector of the normalized adjacency matrix  x = xP  for undirected graphs, the degree distribution 1 2 3 4 5

36. Eigenvalues and Eigenvectors  The value λ is an eigenvalue of matrix A if there exists a non-zero vector x, such that Ax=λx. Vector x is an eigenvector of matrix A  The largest eigenvalue is called the principal eigenvalue  The corresponding eigenvector is the principal eigenvector  Corresponds to the direction of maximum change

37. Types of networks  Social networks  Knowledge (Information) networks  Technology networks  Biological networks

38. Social Networks  Links denote a social interaction  Networks of acquaintances  actor networks  co-authorship networks  director networks  phone-call networks  e-mail networks  IM networks Microsoft buddy network  Bluetooth networks  sexual networks  home page networks

39. Knowledge (Information) Networks  Nodes store information, links associate information  Citation network (directed acyclic)  The Web (directed)  Peer-to-Peer networks  Word networks  Networks of Trust  Bluetooth networks

40. Technological networks  Networks built for distribution of commodity  The Internet router level, AS level  Power Grids  Airline networks  Telephone networks  Transportation Networks roads, railways, pedestrian traffic  Software graphs

41. Biological networks  Biological systems represented as networks  Protein-Protein Interaction Networks  Gene regulation networks  Metabolic pathways  The Food Web  Neural Networks

42. Now what?  The world is full with networks. What do we do with them?  understand their topology and measure their properties  study their evolution and dynamics  create realistic models  create algorithms that make use of the network structure

43. Measuring Networks  Degree distributions  Small world phenomena  Clustering Coefficient  Mixing patterns  Degree correlations  Communities and clusters

44. Degree distributions  Problem: find the probability distribution that best fits the observed data degree frequency k fkfk f k = fraction of nodes with degree k = probability of a randomly selected node to have degree k

45. Power-law distributions  The degree distributions of most real-life networks follow a power law  Right-skewed/Heavy-tail distribution  there is a non-negligible fraction of nodes that has very high degree (hubs)  scale-free: no characteristic scale, average is not informative  In stark contrast with the random graph model!  highly concentrated around the mean  the probability of very high degree nodes is exponentially small p(k) = Ck -α

46. Power-law signature  Power-law distribution gives a line in the log-log plot  α : power-law exponent (typically 2 ≤ α ≤ 3) degree frequency log degree log frequency α log p(k) = -α logk + logC

47. Examples Taken from [Newman 2003]

48. A random graph example

49. Maximum degree  For random graphs, the maximum degree is highly concentrated around the average degree z  For power law graphs  Rough argument: solve nP[X≥k]=1

50. Exponential distribution  Observed in some technological or collaboration networks  Identified by a line in the log-linear plot p(k) = λe -λk log p(k) = - λk + log λ degree log frequency λ

51. Collective Statistics (M. Newman 2003)

52. Clustering (Transitivity) coefficient  Measures the density of triangles (local clusters) in the graph  Two different ways to measure it:  The ratio of the means

53. Example 1 2 3 4 5

54. Clustering (Transitivity) coefficient  Clustering coefficient for node i  The mean of the ratios

55. Example  The two clustering coefficients give different measures  C (2) increases with nodes with low degree 1 2 3 4 5

56. Collective Statistics (M. Newman 2003)

57. Clustering coefficient for random graphs  The probability of two of your neighbors also being neighbors is p, independent of local structure  clustering coefficient C = p  when z is fixed C = z/n =O(1/n)

58. Small world phenomena  Small worlds: networks with short paths Obedience to authority (1963) Small world experiment (1967) Stanley Milgram (1933-1984): “The man who shocked the world”

59. Small world experiment  Letters were handed out to people in Nebraska to be sent to a target in Boston  People were instructed to pass on the letters to someone they knew on first-name basis  The letters that reached the destination followed paths of length around 6  Six degrees of separation: (play of John Guare)  Also:  The Kevin Bacon game  The Erdös number  Small world project: http://smallworld.columbia.edu/index.html

60. Measuring the small world phenomenon  d ij = shortest path between i and j  Diameter:  Characteristic path length:  Harmonic mean

61. Collective Statistics (M. Newman 2003)

62. Mixing patterns  Assume that we have various types of nodes. What is the probability that two nodes of different type are linked?  assortative mixing (homophily) E : mixing matrix p(i,j) = mixing probability p(j | i) = conditional mixing probability

63. Mixing coefficient  Gupta, Anderson, May 1989  Advantages:  Q=1 if the matrix is diagonal  Q=0 if the matrix is uniform  Disadvantages  sensitive to transposition  does not weight the entries

64. Mixing coefficient  Newman 2003  Advantages:  r = 1 for diagonal matrix, r = 0 for uniform matrix  not sensitive to transposition, accounts for weighting (row marginal) (column marginal) r=0.621Q=0.528

65. Degree correlations  Do high degree nodes tend to link to high degree nodes?  Pastor Satoras et al.  plot the mean degree of the neighbors as a function of the degree  Newman  compute the correlation coefficient of the degrees of the two endpoints of an edge  assortative/disassortative

66. Collective Statistics (M. Newman 2003)

67. Communities and Clusters  Use the graph structure to discover communities of nodes  essentially clustering and classification on graphs

68. Other measures  Frequent (or interesting) motifs  bipartite cliques in the web graph  patterns in biological and software graphs  Use graphlets to compare models [Przulj,Corneil,Jurisica 2004]

69. Other measures  Network resilience  against random or targeted node deletions  Graph eigenvalues

70. Other measures  The giant component  Other?

71. References  M. E. J. Newman, The structure and function of complex networks, SIAM Reviews, 45(2): 167- 256, 2003  M. E. J. Newman, Random graphs as models of networks in Handbook of Graphs and Networks, S. Bornholdt and H. G. Schuster (eds.), Wiley- VCH, Berlin (2003).  N. Alon J. Spencer, The Probabilistic Method