Network Theory: Computational Phenomena and Processes Social Network Analysis Dr. Henry Hexmoor Department of Computer Science Southern Illinois University Carbondale Degree, Indegree, Outdegree Centrality Degree Centrality: C D (i ) n x ij j 1 Indegree Centrality: n C I (i ) x ji j 1 Outdegree Centrality: C O (i ) n x i 1 ij C D ( i ) normalized Centrality C D (i ) n 1 Percentage Eigenvector Centrality=CE (i)=I’th entry of eigenvector e ( )V [ 1 , 2 ,..., n ] e = largest eigenvalue of adjacency matrix Betweenness Centrality C B (k ) ikj ij ,i j k ikj= number of geodesic linking across i and j has pass through node k. Normalized betweenness= C B (k ) ( n 1)( n 2 ) 2 Closeness Centrality n C c (i ) d ij j 1 Centrality n : C (i ) j 1 Scaling factor Adjustment ij ( C ( j ) ) One-node network 1-Connection 0-No connection One-mode network: Actors are tied to one another considering one type of relationship; i.e Binary Adjacency matrix v5 v1 v2 v3 v4 v1 0 1 0 1 1 v2 0 0 0 1 0 v3 0 0 0 0 1 v4 0 0 0 0 0 v5 0 1 0 0 0 Two-node network Two-node network: Actors are tied to events. • Incident network • Bipartite graph e.g. Student attending classes Affiliation network Actors are tied to ----Organization/Attributes; e.g. Affinity network, Homophily network Attribute m1………………..Attribute mn Org1……………………..…….Org n 1 2 . . n 1 2 . . n People Sociogram ≡ { } Points-------------individuals Lines --------------relationship Attributes Centrality Star graph A has higher degrees. A is central to all. Centrality: Quantifying a network node. 𝐶𝐷 (i)= 𝑛 𝑗=1 𝑥ij Normalized Centrality E F Centrality: 𝐶𝐷 (A)=4 A D C B Normalized Centrality: 𝐶𝐷 (A) = 4 = 80% ‘ 𝐶𝐷 (A)= 6-1 5 𝐶𝐷 (B)=3, 𝐶𝐷 (C)=2, 𝐶𝐷 (D)=2, 𝐶𝐷 (E)=2 , 𝐶𝐷 (F)=1 A is more central than F Directed network centrality: E F 𝐶𝑜 (A)=3 𝐶𝑜 (F)=1 A D 𝐶𝐼 (A)=2 𝐶𝐼 (F)=∅ C B 𝐶𝐼 (B)=2 𝐶𝐼 (D)=2 𝐶𝐼 (E)=2 𝐶 (A) 2 𝐷 𝐶𝐷‘(A)= = = 40% (Normalize centrality) 5 6-1 Prestigue of A=B=C=E=2 (Indegree) Eigen vectors Vector X is a matrix with n rows and column, linear operator A, maps the vector X to matrix product AX 𝑥1 𝑥2 . 𝐴1,1 𝐴2,1 . . → . . . . . 𝑥𝑛 𝐴𝑛,1 𝐴1,2 … … . 𝐴2,2 … … . . . . . 𝐴𝑛,2 → A𝑋 = ƛ. A 𝐴1,𝑛 𝑥1 𝐴2,𝑛 𝑥2 . . . . . . . . 𝑎𝑛,𝑛 𝑥𝑛 = 𝑦1 𝑦2 . . . . 𝑦𝑛 Eigen Value 3 −3 ƛ to the equations |𝐴 − ƛ. I| ≠ ∅ Second degree centrality Node A Value .534 B .275 C .363 D .199 E .199 F .199 G .199 H .102 I .102 Eigen value centrality J .164 𝐶𝐷 (A)= 𝐶𝐷 (O)=6 higher Eigen values K .164 L .164 M .164 N .164 O .441 Consider this 16 degree graph network: I H B M N O C G A D E L K J F Betweenness Centrality Betweennes centrality measures the extent to which a vertex lies on paths between other vertices. 𝛿𝑖𝑘𝑗 =Number of paths from i to j passing through k 𝐶𝐵 (K)= 𝛿𝑖𝑘𝑗 𝛿𝑖𝑗 𝛿𝑖𝑗 = Number of shortest distance path from i to j K= Geodesic distance Normalized centrality: E A F D Node No of distinct path from the node Normalized (C’) E 4.0 40% A 3.5 35% D 1.0 10% B 0.5 5% C 0.0 0% F 0.0 0% C B 𝐶𝐵 (𝐾) 𝐶′𝐵 = [(𝑛−1)(𝑛−2)/2] Closeness Centrality Closeness centrality is the mean distance from a vertex to other vertices. E F A D C Node (i) 𝐶𝑓 𝐶𝐶 𝐶𝐶 (%) B A 6 5/6 83% E 7 5/7 72% D 7 5/7 72% B 8 5/8 63 % C 9 5/9 55% F 11 5/11 46% 𝑛 𝑗=1 𝑑𝑖𝑗 𝐶𝑐 (i)=[𝐶𝑓 (i)]−1 𝐶𝑓 (i)= f= farness c= closeness d= distance between i & j n= total number of nodes x (𝑛 − 1) Eigen vector Eigen vector for N(i) = Neighbours of i = {J ∈ 𝑉 | 𝑖, 𝑗 ∈ E} where N=(𝑉, ∈) Eigen vector centrality: 𝐶𝐸 (i)= J∈𝑁(𝑖) 𝐶𝐸 (J) Therefore, ƛ. E = AT . E Page Rank Cetrality The numerical weight that it assigns to any given element E is referred to as the PageRank of E and denoted by PR(E). Page Rank Centrality: 𝐶𝑃𝑅 (i)= 𝐶𝑃𝑅 (J) 𝐽∈𝑁(i) 𝑑𝑒𝑔 𝑂𝑢𝑡 (J) Bonacich/Beta Centrality • Both centrality and power were a function of the connections of the actors in one's neighborhood. • The more connections the actors in your neighborhood have the more central you are. • The fewer the connections the actors in your neighborhood, the more powerful you are. • It is the weighted centrality Bonacich/Beta Centrality: 𝑪𝑩𝑪 = 𝑱∈𝑵(𝒊)(𝜶 + [𝜷 x 𝑪𝑩𝑪 (𝐉)]) =𝜶 𝐥𝐨𝐠 𝒏 𝐢 + 𝜷. 𝑱∈𝑵(𝐢) 𝑪𝑩𝑪 (𝑱) Here, 𝛼 = Importance of the node i′ s degree (local importance) 𝛽 = Imporatance of neighbour ′ s centrality (global imporatance) Density Density: It is the level of ties/connectedness in a network; It is a measure of a network’s distance from a complete graph. Complete graph: Every node is connected to every node in the network Ego Density L = number of links in network n = number of nodes in the network Structural Hole (Ron Burt) Let’s consider this, 1 gap 2 Structural Hole • The gap between connected components is the hole • Structural hole provides diversity of information for nodes that bridge them • Without structural hole information becomes redundant and less available Brokering Brokering is bridging different group of individuals. 1.Coordinator (local brokers; Intragroup brokering) C e.g. manger, mediating employees B A B as Coordinator/ Broker 2. Consultant (Intergroup brokering by an outsider e.g. middle man in business between buyers &seller, stock agent ) Consultant Seller Buyer Brokering 3. Representation (represents A when negotiating with C) e.g. hiring a mechanic to buy car for you A C B 4. Gate Keeper(e.g a butler, chief of staff) Actor Producer Agent Actor Producer Dyadic Relation Dyads: A B A B A B A B Triads: when a triad consists of many ties, an open triad (triangle) is forbidden. A C B0 Triad Relations (census) Components Component is a group where all individuals are connected to one another by at least one path. • Weak Component: A component ingoing direction of ties. • Strong Component: A component with directional ties. • Clique: A subgroup with mutual ties of three or more. who are directly connected to one another by mutual ties Bonacich Centrality o CBC = Degree Centrality High Degree + Low Betweenness : Ego Connection are redundant Low Closeness+ High Betweenness : Rare node but pivotal to many In triads, there is a structural force toward transitivity. Bonacich Principle of strength of weak tie.(Granovetter, 1973): There is a social force that suggests transitivity. If A has ties to B and B to C, then there is tie from A to C. Diameter Reverse distance: Max ( d ij ), i , j V RD ij d ij (1 Diameter ) Integration: RD I (k ) Reverse distance jk jk n 1 Network distance I (k ) I ( k ) max jk ( RD jk ) Degree to which a node’s inward ties integrate it into the network. Radiality Degree to which a node’s outward ties connects the node with novel nodes. RD R (k ) kj jk n 1 R (k ) R ( k ) max jk ( RD jk ) Edge between-ness Number of shortest path from s to t that pass through edge e C EB ( e ) st ( e ) st Number of shortest path from s to t This is important in diffusion studies like epidemics Social Capital The network closure argument: Social Capital is created by a strongly interconnected network. The structural hole argument: Social Capital is created by a network of nodes who broker connections among disparate group. Structural Equivalence= similarity of position in a network Euclidean Distance [( x ik x jk ) ( x ki x kj ) 2 2 i j k E.g., A A B 0 0 C D 0 E 0 B C D E 0 1 1 1 1 0 0 0 1 0 0 0 0 0 d 1 AB 0 0 E B A C D A B have distance zero d DE 2 1 . 41